The reason I bring this up is that I saw a post from BeauHD on Slashdot last week claiming that Emacs is over fifty years old. As one of the resident greybeards in the infosec industry (I’m officially “Get-of-my-lawn!” old), I have a bit of nostalgia for all the old things computer nerds like me used before the internet became a thing (pre-1990s).

Emacs: What is it?

For those who have no idea what I’m talking about, Emacs was, and is, a text editor, but it’s unlike any editor you have ever used. It’s highly customizable. You can assign complete programs to every keypress and/or keypress combination. In the nerd world, we call those customizations macros. Emacs stands for Editing Macros.

Unix wizards built it on top of its own programming language called Emacs Lisp, which let users build email readers, calendars, file managers, even games. Try to do that with Google Docs or Microsoft Word. The old joke floating around on bulletin boards at the time was that “Emacs is a great operating system that just lacks a good text editor.” (See what they did there? That’s nerd humor at its finest.)

The Hatfields and the McCoys: Emacs vs vi

When I was in grad school back in the late ‘80s, before there was an internet really, before the World Wide Web, before AOL, Unix wizards fell into two camps about which text editor to use: Emacs vs vi.

vi (and its modern version, Vim) is like a pocketknife.

• It’s small, it’s fast, and it comes on almost every Unix computer in the world.

• It has “modes.” In one mode, your keys type letters. In another mode, the same keys give commands: d means delete, w means word, so dw deletes a word.

• Newcomers get stuck because they don’t know how to leave. “How do I exit vi?” is one of the most-asked programming questions ever. (Answer: press Esc, then type :q!.) Easy to remember, right?

And not to get meta or anything, some Emacs users have a macro that lets them run vi inside of Emacs, because, you know, Unix wizards have nothing better to do.

The bottom line is that vi is about editing fast and Emacs is about creating an entire editing environment that best suits your needs. Both have outlasted dozens of “modern” editors.

There’s an entire history of how Richard Stallman wrote GNU Emacs in the mid-1980s and gave it away with its source code. It started the GNU Project and the GPL license, which later became the foundation for Linux.

My Love-Hate Relationship with Emacs

I was in grad school at the time (late 1980s) assigned to a group capstone project. I forget what the actual project was about, but it involved turning in a 100-page architectural report with fancy formatting and footnotes.

Point-and-click user interfaces were around, but they weren’t popular yet. So you drove Emacs with the keyboard, using key combinations like Ctrl-x Ctrl-s to save a file. Emacs fans loved it, but newcomers found it baffling. I know I did and that’s why I have a love-hate relationship with it.

Back then, no editors had a WYSIWYG user interface (What You See Is What You Get). To do the formatting for a hard-copy printout, Unix wizards cast esoteric printer spells on the text document using programs like LaTeX, Scribe, or troff, and sent it to a printer. If you didn’t like the way it looked after you printed it, you recast the spells and sent it to the printer again. You can imagine the number of trees we killed going through drafts. I’m pretty sure I was at least partially responsible for the deforestation of the lower Amazon region.

After several all-nighters before the deadline, we were about done. My group assigned me the task of taking one last grammar and spelling look before we turned it in. (I’m not sure why I got this task. Maybe it was because my contribution to the project at this point wasn’t measurable).

But remember when I said users could write macros for every key combination they could think of? Well, I don’t know how I managed it, but I accidentally found the macro that capitalized every letter in the document. Let that sink in. I didn’t just capitalize all the words that made up the report; I also capitalized all of the LaTeX incantations, the markups that made the formatting work, which was all case sensitive. In fact, most LaTeX commands are lowercase. Capitals usually mean something specific. Now they were all capitalized. Doh!

My immediate first thought was, why would you even have a macro for that!?! I mean, even if you assume that you might have to capitalize every word in a document, say every 10 years or so, why would you spend time automating that task? That guy (and it was most assuredly a guy) needed to get a life.

My second thought was: what is the Emacs macro for undo? You, dear reader, might be thinking this isn’t a big problem because all Unix Wizards past the 3rd or 4th level know that saving a file with Ctrl-u Ctrl-x Ctrl-s forces a backup copy. I could just reload the previous version. But I was just an apprentice and didn’t know that, nor did any of my fellow group neophyte apprentices. We weren’t saving the file that way.

Emacs does have an undo macro (Ctrl-x u), but it was so non-intuitive, that when I tried to use it, I destroyed all previous versions of the document. I know! It shouldn’t have allow me to do that. The only copy that I had was the report that was fully capitalized. Needless to say, I was, and still am, 30 years later, the most popular guy from our project group.

My punishment was to spend a couple more all-nighters retyping that 100-page report. But this time, I asked a wizard for help. Looking down his nose at me with an “I can’t believe you don’t know this” look, he showed me how to set up the Ctrl-u Ctrl-x Ctrl-s incantation.

Takeaway

So happy birthday, Emacs, you magnificent, infuriating spellbook. Fifty years on, you’re still proof that Unix wizards can build an entire universe inside a text editor and forget to label the emergency exit. I admire the power, the customization, and the give-it-away spirit that helped generations of nerds build things of their own. I also remember that putting unlimited magical power at an apprentice’s fingertips can turn a hundred-page report into a hundred-page scream. My fellow apprentices eventually graduated, the Amazon is hopefully recovering, and I learned that the most important spell in any wizard’s repertoire is “make a backup.” Preferably before the idiot in the pointy hat touches the keyboard.

Source

BeauHD, 2026. EMACS Turns 50 [News]. Slashdot, URL: https://news.slashdot.org/story/26/10/28/1541214/emacs-turns-50

As I write this, Halloween is less than a week away. In anticipation of my favorite holiday, this month, I’ve been reviewing some new and old-favorite vampire novels. I’ve done two so far:

• Keith Rosson’s 2026 Coffin Moon. (https://rebrand.ly/Howard-Review-Coffin-Moon)

• Bram Stoker’s 1897 Dracula. (https://rebrand.ly/Howard-review-dracula)

For the last one in the series, I’m reviewing my favorite vampire novel of all time: ‘Salem’s Lot. And not just the novel, I’m covering all three TV adaptations too, because I can’t help myself.

I’ve been a fan of scary stories since I was a wee lad. But, in 7th grade, when I saw the ABC TV Movie of the Week The Night Stalker, I became an instant vampire fan. I still think it’s the best vampire movie ever made (and you can watch it on YouTube for free). I re-watch it every halloween. I just watched it last week. Fantastic.

So, of course, when I found Stephen King in the 1980s, I was hooked. I’ve been reading him ever since. As of October 2026, I’ve read 38 out of his 83 published books. And, because of the 2024 HBO movie adaptation, I read ‘Salem’s Lot one more time last year.

Vampire Novels and Stephen King

And I love me some vampire books. Book sellers have published almost 100 vampire novels over the years and I track them in a Spreadsheet. I’ve read 18 of them so far. ‘Salem’s Lot is not only my favorite Stephen King story but also my favorite vampire novel, just ahead of John Steakley’s Vampire$, Bram Stoker’s Dracula, and Laurell K. Hamilton’s Guilty Pleasures. Of the King books I’ve read, Pet Sematary is the scariest, but ‘Salem’s Lot was my first King novel and the one that hooked me.

In the book’s prologue, King tells a lovely story about his mother when he and his brother were kids. She let them read anything they wanted when they were probably too young. She would always give her review before handing the library book over. “This is trash.” Or, “This is bad trash.” She died before King published ‘Salem’s Lot many years later, but he always liked to think that she might have read the novel in one of her famous, middle-of-the night, marathon reading sessions. Afterwards, she would pronounce it as trash, but not bad trash.

I love that story.

Horror might not be your thing, but there’s no denying that King has a remarkable talent for telling popular tall tales. I asked Claude to estimate the volume of Stephen King’s work:

• 80 to 85 “published books” (counting novels, collections, and nonfiction as separate titles)

• 350 to 400 million books sold

• 70 to 100+ movie and TV adaptations (the range depends on how spinoffs and miniseries are counted)

The man is prolific and he knows how to write. He wrote a nonfiction book back in 2002 on his writing process, called On Writing, and I often recommend it to colleagues and subordinates looking to improve their writing game.

One reason for his success is that King has the ability to create deeply engaging narratives that are accessible to the average person. His dialogue is conversational. His stories are character driven, not plot driven. He is infamous for ignoring plot as a way to tell a story. He simply drops interesting characters into a crazy situation (Mom and kid trapped in a car by a rabid dog) and writes what happens next. One common criticism though is that he sometimes fails to stick the landing, mainly because he doesn’t know what the ending is when he drops his people into a crazy situation.

For his movies, in my humble opinion, each adaptation generally falls into one of four buckets of goodness:

• Good Movies that happen to have a horror theme (16% - Like the Green Mile)

• Good Horror Movies (32% - Like Carrie, The Shining)

• Trash Horror Movies (32% - Like Hearts in Atlantis, Needful Things)

• Bad Trash Horror Movies (20% - Like The Dark Tower, Silver Bullet )

That’s a pretty good track record. That said, there have been three attempts to bring ‘Salem’s Lot to the screen. I would put them all in the trash horror movie bucket, but not bad trash.

Spoilers

‘Salem’s Lot, the novel, begins with Ben Mears, a slightly famous author (maybe a King stand in?), returning (after 25 years) to his childhood hometown of Jerusalem’s Lot (nicknamed ‘Salem’s Lot) . His intent is to write about the old house on the hill, the perhaps haunted Marsten House. As a kid on a dare, he experienced something terrible there. The story itself covers just over 30 days from the time Ben (as an adult) arrives in town until the climax of the story.

When Ben arrives, he discovers that two businessmen, Barlow and Straker, who are opening an antique shop in town, had just bought the house. We find out later that Barlow is the vampire and Straker is the familiar. Ben befriends a local high school teacher Matt Burke and begins a romantic relationship with Susan Norton, a young college graduate. After Ben settles in, a young boy, Ralphie Glick, disappears and his brother, Danny, dies mysteriously. At this point, the reader suspects that Barlow is actually an ancient vampire, and Straker is his human servant. Ralphie Glick becomes the town’s first vampire, turned by Barlow.

King says that he read Bram Stoker’s Dracula when he was very young and loved it. When he read Tolkien’s Rings Trilogy ten years later, he made the stunning observation that the story was “a slightly sunnier version ... with Frodo playing Jonathan Harker, Gandalf playing Abraham Van Helsing, and Sauron playing the Count himself.” Ever since I learned that, I’ve been calling the Lot’s gang of misfit heroes, the fellowship. Almost all are tokens of Stoker’s characters too, not exact replicas per se but more like you can see the resemblance if you squint at the right moment when the sunlight is perfect.

As Barlow turns more townspeople, Ben (similar to Dracula‘s Jonathan Harker) forms a group (The Fellowship) to combat the threat including Susan (Mina Harker), Matt (Van Helsing), a local doctor named Jimmy Cody (Dr. John Seward), young Mark Petrie (friend of Ralphie and Danny, a King original character who could be an amalgam of the Texan Quincey Morris, Van Helsing, Mina Harker, and Lucy Westenra), and Father Callahan (the local Catholic priest, another King original).

Riffing on Shirley Jackson’s The Haunting of Hill House, King introduces the idea that a place can be a kind of battery for evil, an absorber of evil things long after the perpetrators have left. In the Lot’s case, it’s the Marsten House. King has returned to that idea in The Shining and IT to one degree or another. Barlow (Dracula) is drawn to the house as he moves from Europe to establish his home base in America.

King’s retelling of the Stoker vampire story in a modern America was radical at the time. Today in modern culture, vampires are everywhere. But back then, having Dracula show up in your local home town was a brand new thing, an original idea, and scary as hell.

But he inverts the Stoker novel. The heroes in the Dracula novel seem destined to win with their use of science, modern day technology tools, and their belief that good will conquer evil. The fellowship, on the other hand, seems impotent. None of their modern day conveniences seem to be any use at all. In desperation, they reach for the power of the old Catholic church (which fails big at least once). They seek wisdom from old historic tomes. Their weapons are wooden stakes and holy water. And the ‘Salem’s Lot people swim in a soup of despair, not jumping into a full throated good vs bad battle, but just trying to survive in a fog of benign evil. And even though I said that Pet Sematary was King’s scariest, ‘Salem’s Lot has some intense scary scenes too.

The obvious solution to this madness is for the Fellowship to march up to the Marsten house, during the day, and drive a wooden stake through Barlow’s heart. They know they have to do that but what I love about the story is that King finds imaginative ways to prevent the fellowship from doing that specific thing. They need stakes (where do you find enough wooden stakes to kill an entire town because day-by-day, the town is filling up with vampires?), they need holy water, they need a priest, they get separated, they are across town, and everything eats the clock. As the fellowship sets off to accomplish these things, Barlow anticipates their every move. One by one, they get picked off. They are delayed and outmaneuvered at every turn and the clock keeps ticking.

Criticisms

As much as I do love this book, I have a couple of small nit picks. First is that King is juggling many characters. He explores the slow infestation of the town across those 30 days and he does that by taking several tours around town. The first time through is the introduction to the normal blue collar townspeople, their troubles and tribulations. These are real people. We like them, well, most of them. The second rotation is to check in on how everybody is doing. There have been several reported deaths and many are out sick. People are scared. And the last rotation is the capitulation, the result of the town becoming completely overrun. That’s all great storytelling but it does take time and the reader has to keep track of all those characters. Just when we want to get to the main action, King tells us another story about one of the townies. It’s frustrating.

My second criticism is that King leans on the same trope every cheap B-horror movie uses: someone does something stupid and walks right into the monster’s trap. The usual suspects:

• Running upstairs instead of out of the house.

• Going to the cemetery at night instead of during the daytime.

• The group splitting up instead of sticking together.

Susan falls into this one. She doesn’t fully buy the vampire story the boys are pushing, so she decides to drive up to the vampire house (the one everybody in town already thinks is haunted) by herself, to have a stern talking-to with Mr. Barlow and Mr. Straker, thank you very much, and get some answers.

The TV Adaptations

1979

Amazingly, there have been three tries to recreate the magic of the book for TV. The first was the 1979 mini-series directed by Tobe Hooper of The Texas Chain Saw Massacre and Poltergeist fame. It starred David Soul as Ben Mears. Soul was red hot as a TV actor at the time coming off the lead for the TV series Starsky and Hutch, one of the most popular TV shows in the late 1970s. His love interest, Susan Norton, was played by Bonnie Bedelia, famous for the Die Hard movie franchise and the TV show Parenthood. It also starred James Mason, the classically trained English actor famous for Alfred Hitchcock’s North by Northwest and Disney’s 20,000 Leagues Under the Sea. He played Barlow’s familiar: Straker.

For a 1970s, TV mini-series, it was pretty scary and it’s probably the best of the three. It didn’t quite capture the camaraderie of the Fellowship though. I guess it didn’t help that David Soul wasn’t a great actor. The director, Hooper, for some reason, also decided to change the look of the vampire Barlow. From the book’s description, he is a handsome and debonare senior eastern European (like when Dracula arrives in London). But Hooper changed him to look like the vampire in the 1922 movie, Nosferatu: A Symphony of Horror. I guess he thought he needed it to be more scary. It was an odd choice and gave the impression that the familiar, Straker, was in charge and not the other way around, which is a complete and unnecessary departure from the book. Also, in this version, Ben doesn’t kill Susan at the Marsten house. He can’t bring himself to do it. When he and Mark run away to Mexico after killing Barlow, Susan tracks him down and then Ben stakes her. It’s an interesting idea and would have made a great sequel to the book (Ben and Mark on the run from the surviving ‘Salem’s Lot vampires).

2004

25 years later (2004), the director Mikael Salomon (a journeyman TV director at the time) gave it a second try with Rob Lowe playing Ben Mears. Lowe was famous for being part of the brat pack and was midway through his run as Sam Seaborn in The West Wing TV Show. The rest of the cast is outstanding too. Donald Sutherland (Famous for The Dirty Dozen, Mash the movie, Ordinary People, JFK, the movie version of Buffy the Vampire Slayer, and The Hunger Games), played a deliciously evil Straker. And Rutger Hauer (famous for Blade Runner and was also in the Buffy movie with Sutherland) played a good looking vampire (Barlow) in this version, more akin to how King described him in the book. Andre Braugher (famous for Glory, The Mist, Men of a Certain Age, and Brooklyn Nine-Nine) plays the “gay” English teacher, Matt Burke.

When I read the book the first time in the 1980s, I may have been naive, but I never registered Matt as being gay. He is a confirmed bachelor for sure but he never showed any overt signs. In this version, when Mike comes back to Matt’s house after being dead, he plays the gay card telling Matt that he always knew that Matt was interested as a way to convince Matt to come to the other side.

Callahan is played by the great character actor James Cromwell, best known for Babe, L.A. Confidential, and The Green Mile. He’s fantastic here. But Salomon, the director, changed the ending in this version. When Callahan takes the dark communion from Barlow, the director implies, but never really shows, that Callahan replaces Straker as Barlow’s familiar. After Ben and Mark torch ‘Salem’s Lot, they track Callahan to a soup kitchen in NYC and have a knock-down drag-out fight where both Callahan and Ben die. That was odd and unnecessary. But it also feels like he edited out an entire section that might have explained that better.

The 2004 version is dark and grim. There is no humor in it whatsoever. It’s not very enjoyable.

2024

In 2024, Director Gary Dauberman tried again. He is best known for the Annabelle and Nun franchises plus the remake of another King movie adaptation IT (Chapter 1 and Chapter 2). Lewis Pullman, famous for 2023’s Lessons in Chemistry and 2022’s Top Gun: Maverick, plays Ben. Alfre Woodard, the perennial “That Guy” actress, plays the female Dr Cody. Pilou Asbæk (best known for Game of Thrones) plays Straker.

This version is a good popcorn movie with some legitimate good scares and ideas. At one point, in the middle of the night, Ben and Cody race Susan to a church in an attempt to prevent her from turning. She escapes and Ben chases her out into the empty streets. Cody yells that it’s not safe. Ben turns to look at her and then looks at the street again. At least 30 vampires have appeared on the building roofs and the aforementioned empty streets. Ben sprints back to the safety of the church as a flying vampire is chasing him. Cody tosses him a crucifix in the nick of time (or a vial of holy water - I couldn’t tell) just as the vampire grabs Ben and lifts him into the air. From Cody’s perspective inside the church, she sees them both land hard at the door’s front steps with the Vampire’s blood spilling everywhere (Not quite sure what happened there - but it’s a great horror scene). The other great idea is that, as more and more ‘Salem’s Lot citizens turn into vampires, the places they choose to hide in the daytime are the trunks of cars abandoned at the local drive in theater. That is inspired. But the movie itself is a shadow of the book. It’s like a xerox copy of a xerox copy of a xerox copy.

Take Away

With all of that said, I think that King’s mom would have rated all three efforts as Trash Horror movies, not Bad Trash Movies.

But, ‘Salem’s Lot, the novel, is a treasure. I’m giving it a five-star rating for myself which I don’t give out that much. It means that I will recommend this book to others, not just horror fans, and that I know I will re-read it sometime in the future.

Sources

Stephen King, 1975. ‘Salem’s Lot [Novel]. Goodreads, URL: https://www.goodreads.com/book/show/11590._Salem_s_Lot

Tobe Hooper (Director), Davis Soul, Bonnie Bedelia, Fred Willard (Actors), 1979. Salem’s Lot [Movie Social Media Entry]. Letterboxd, URL: https://letterboxd.com/film/salems-lot/

Mikael Salomon (Director), Rob Lowe, Andre Braugher, Donald Sutherland. Rutger Hauer, and James Cromwell (Actors), 2004. Salem’s Lot [Movie Social Media Entry]. Letterboxd, URL: https://letterboxd.com/film/salems-lot-2004/

Gary Dauberman (Director), Lewis Pullman, Alfre Woodard (Actors), 2024. ‘Salem’s Lot [Movie Social Media Entry]. Letterboxd, URL: https://letterboxd.com/film/salems-lot-2024/

References

Alisha Grauso, 2024. All 16 Stephen King Books That Tie Into & Reference ’Salem’s Lot [Explainer]. Screen Rant. URL https://screenrant.com/salems-lot-town-events-story-stephen-king-books-tie-ins-appearances/

Bram Stoker, 1897. Dracula [Book]. Goodreads, URL: https://www.goodreads.com/book/show/17245.Dracula

Eric Eisenberg, 2024. Max’s Salem’s Lot Review: The New Adaptation Squeezes Stephen King’s Story But Is Sustained By Its Characters And Scares [Review]. CINEMABLEND. URL https://www.cinemablend.com/movies/salems-lot-review

F.W. Murnau, 1922. Nosferatu: A Symphony of Horror [Movie]. IMDb. URL https://www.imdb.com/title/tt0013442/fullcredits

Gary Dauberman, 2024. Salem’s Lot [Movie]. IMDb. URL https://www.imdb.com/title/tt10245072/

Gem Wheeler, 2024. Kolchak: The Night Stalker at 50 - A Character Who Changed the Horror Genre Forever [Essay]. Den of Geek. URL https://www.denofgeek.com/tv/kolchak-the-night-stalker-at-50-a-character-who-changed-the-horror-genre-forever/

Jason Katims, 2010. Parenthood [TV Series 2010 - 2015]. IMDb. URL https://www.imdb.com/title/tt1416765/

John Llewellyn Moxey, 1972. The Night Stalker [TV Movie]. IMDb. URL https://www.imdb.com/title/tt0067490/

John McTiernan, 1988. Die Hard [Movie]. IMDb. URL https://www.imdb.com/title/tt0095016/

Keith Rosson, 2025. Coffin Moon [Book]. Goodreads, URL: https://www.goodreads.com/book/show/222419255-coffin-moon

Padraig Cotter, 2019. All Stephen King Movies Ranked, From Worst To Best [Essay]. Screen Rant. URL https://screenrant.com/all-stephen-king-movies-ranked/

Richard Fleischer, 1955. 20,000 Leagues Under the Sea [Movie]. IMDb. URL https://www.imdb.com/title/tt0046672/

Rick Howard, 2026. Coffin Moon Review: A Vampire Story to Set this Year’s Halloween Mood: Gritty, Bloody, and Full of Throat-Ripping Monsters [Book Review]. Rick’s First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-Review-Coffin-Moon

Rick Howard, 2026. Dracula Review: Still Scary After 129 Years [Book Review]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-review-dracula

Shirley Jackson, 1959. The Haunting of Hill House [Book]. Goodreads. URL https://www.goodreads.com/book/show/89717.The_Haunting_of_Hill_House

Shmoop Editorial Team, 2008. ‘Salem’s Lot Introduction [Review] Shmoop. URL: https://www.shmoop.com/study-guides/salems-lot/

Smythe, James. “Rereading Stephen King: Week Two – Salem’s Lot.” the Guardian. The Guardian, June 11, 2012.

Staff, 2026. 10 Greatest Vampire Movies of All Time [Analysis]. Backyard Movie Critic (Substack), URL:

Backyard Movie Critic

10 Greatest Vampire Movies of All Time

Vampire movies are kind of the backbone of horror. Throughout the 1930s–1960s, horror was all about dark, creepy castles and pale-skinned, red-eyed bloodsuckers. Vampires represent death, sex, and even the fear of disease. They represent deep, primal fears. Honestly, they go along with my theory that horror is the only genre that appeals to one of our p…

Read more

17 days ago · 119 likes · 96 comments · Backyard Movie Critic

Stephen King, 2000. On Writing: A Memoir of the Craft [Book]. Goodreads. URL https://www.goodreads.com/book/show/10569.On_Writing

Stephen King, 1983. Pet Sematary [Book]. Goodreads. URL https://www.goodreads.com/book/show/33124137-pet-sematary

Steven Spielberg, Tobe Hooper, 1982. Poltergeist [Movie. IMDb. URL https://www.imdb.com/title/tt0084516/

Tobe Hooper, 1974. The Texas Chain Saw Massacre [Movie]. IMDb. URL https://www.imdb.com/title/tt0072271/

William Blinn, 1975. Starsky and Hutch [TV Series 1975 - 1979]. IMDb. URL https://www.imdb.com/title/tt0072567/

I used Claude to

• Check for grammar and spelling errors.

• Flag passive voice.

• Suggest changes to awkward sentence construction.

• Create images.

• Fact-check

• Stress-test my thesis

In the spirit of Halloween, I’ve been reviewing vampire novels. Last week, I reviewed
Keith Rosson’s Coffin Moon.

Review: Coffin Moon

This week, I’m revisiting the original, the 1897 classic, Bram Stoker’s Dracula. This is the big daddy vampire novel although it’s not the first horror novel nor is it the first vampire story.

Milestones in Gothic, Horror, and Vampire Fiction.

But it may be the most influential in terms of worldwide cultural impact. The book has been in print since its first publication in 1897 and creators have made countless versions of it in movies, television, radio, and comic books.

The writing style for the original is a bit dated for modern readers’ tastes. Many novelists back in the day tended to spend a lot of time explaining and describing things. This tends to bog down the pace that is expected from modern readers. Dracula suffers from this a bit.

But I want to make the case here that Dracula has some jaw-dropping-scary scenes; especially when you consider 1897 audiences reading it for the first time.

The book is an epistolary novel meaning that Stoker tells the story entirely through letters, diary entries, other documents, and even an Edison-style cylinder phonograph. The term “epistolary” comes from the Latin word “epistola,” which means “letter.” Although jarring if you’ve never experienced it before, the method lends itself to credibility. It’s similar to found footage films like The Blair Witch Project in the modern day. Readers feel like they are right there in the action with the heroes. The first time I remember seeing it was in Stephen King’s Salem’s Lot. He used newspaper accounts at the front of several chapters as a kind of exposition mechanism that made the story feel more real.

But it does make it difficult for the reader to keep track of everything. According author R. Shanon, it’s just confusing. After reading it several times, she says that “Reading through the letters as a reader, it is enjoyable and they do add to the element of suspense and a low-grade horror that continues through the book. However, after reading the book several times, I would never have been able to give an account of who everyone was and where they all went in the story. That was all too elusive with going back and forth between the letters from different people and then back to the narrative itself.”

I concur and I had a better experience with the audible version, especially with first class voice talents like Alan Cumming and Tim Curry playing key parts. I still struggled with what actually happened though. As best as I can piece together, this is the sequence (Heavy Spoilers follow):

And for those who need a scorecard to keep everybody straight, here you go:

This is the third time I’ve read this book, but the first time I’ve noticed how often Dracula operates offstage. He dominates Harker’s opening account, then largely recedes while the other characters piece together what he’s doing. There is no Dracula journal, no direct access to his thoughts. We know him through his words, his victims, and the evidence he leaves behind.

Stephen King uses a similar approach in ’Salem’s Lot, his homage to Stoker’s novel. Barlow makes appearances before the climax, but King keeps him largely in the background while his influence spreads through the town. In both books, the vampire’s presence is much larger than the space he occupies on the page.

Of course, this novel is the blueprint for all subsequent vampire stories. Every vampire story that comes after has some version of these rules.

One note, in this story, sunlight doesn’t kill Count Dracula; it just diminishes his supernatural powers.

Take Away

On my third trip through Dracula, I still had to work to keep the characters and chronology straight, but the scares held up. The pacing asks for patience; the Count crawling down the castle wall rewards it. Stoker lets us assemble the evidence alongside his vampire hunters, making Dracula’s presence felt even when he’s nowhere on the page. If you love vampire stories, especially ’Salem’s Lot, this classic deserves your time. The audiobook helped me navigate the journals and letters, and beneath the Victorian prose I found a monster who can still make me want to sleep with the lights on.

Source

Bram Stoker (author), Alan Cumming, Tim Curry, Simon Vance, Susan Duerden, John Lee, Graeme Malcom, and Steven Crossley (Narrators), 1897. Dracula [Book]. Goodreads, URL: https://www.goodreads.com/book/show/17245.Dracula

References

Caleb Foster, 2025. Book review: Dracula by Bram Stoker [Analysis]. Vocal, URL: https://vocal.media/bookclub/book-review-dracula-by-bram-stoker

Hinson, H., Nov 1992. “Bram Stoker s Dracula” [Book review]. The Washington Post. URL https://www.washingtonpost.com/wp-srv/style/longterm/movies/videos/bramstokersdracularhinson_a0a7c7.htm.

Horace Walpole, 1764. The Castle of Otranto [Book]. Goodreads, URL: https://www.goodreads.com/book/show/12923.The_Castle_of_Otranto

Matthew Gregory Lewis, 1796. The Monk [Book]. Goodreads, URL: https://www.goodreads.com/book/show/122780618-the-monk

McGeorge, L., 2015. The History of the Horror Novel [Website]. Horror Novel Reviews. URL https://horrornovelreviews.com/2015/09/24/the-history-of-the-horror-novel/

Milo, 2014. Dracula by Bram Stoker - review [Book review]. The Guardian. URL https://www.theguardian.com/childrens-books-site/2014/feb/04/review-dracula-bram-stoker

Njoku, I., 2022. Dracula Review [Book review]. Book Analysis. URL https://bookanalysis.com/bram-stoker/dracula/review/

Ransley, L., 2011. Dracula by Bram Stoker – review [Book review]. The Guardian. URL https://www.theguardian.com/books/2011/apr/17/bram-stoker-dracula-review-classics

Rick Howard, 2026. A Vampire Story to Set this Year’s Halloween Mood: Gritty, Bloody, and Full of Throat-Ripping Monsters [Book Review]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-Review-Coffin-Moon

R. Shannon, July, 2024. Book Review: Dracula by Bram Stoker [Analysis]. Goodreads, URL: https://www.goodreads.com/author_blog_posts/24938619-book-review-dracula-by-bram-stoker?tab=book

Sangrey, K., 2016. Dracula Timeline [Website]. prezi.com. URL https://prezi.com/dciuuf3snvuj/dracula-timeline/

Staff, 2025. ‘Spine-tingling’ stories in the ‘blood-curdling’ British Library [Website Article]. British Library, URL: https://www.bl.uk/stories/blogs/posts/spine-tingling-stories-in-the-blood-curdling-british-library

I used Claude Code to

• Check for grammar and spelling errors.

• Flag passive voice.

• Suggest changes to awkward sentence construction.

• Create images.

• Fact-check

• Stress-test my thesis

I used ChatGPT to make the posters.

Every Halloween, one of my daughters, my favorite niece, and I choose a horror book and read it together. The rest of my family members (my wife, my son, and my other daughter) don’t participate because they don’t particularly like the horror genre so, as you would expect, I have written them out of the will. I’m pretty sure I’ve never really liked them anyway.

This years’ selection is the Keith Rosson vampire book, Coffin Moon.

And just let me say, I love vampire books. I’ve read a bunch of them (although I have many more to get to). Years ago, I started shuffling my personal favorite vampire novel top 10 list, because, you know, I’m a nerd and that’s what nerds do. Every time I read another vampire book, I judge whether it cracks my top 10. Before Coffin Moon, this is how the list stood:

Howard’s Personal Top 10 Vampire Novels

1: Salem’s Lot by Stephen King

2: Vampire$ by John Steakley

3: The Kolchak Papers by Jeff Rice

4: Interview with a Vampire by Anne Rice

5: Dracula by Bram Stoker

6: Let the Right One In by John Ajvide Lindqvist

7: The Southern Book Club’s Guide to Slaying Vampires by Grady Hendrix

8: The Lesser Dead by Christopher Buehlman

9: The Buffalo Hunter Hunter by Stephen Graham Jones

10: Nightwatch by Sergei Lukyanenko

The question is, will Coffin Moon crack the Top 10. Before I decide, let me just say that this book is my kind of vampire story: gritty, bloody, and full of throat-ripping monsters. It’s set in the 1970s so the author doesn’t have to contrive a plot point where the cell phones aren’t useful. There are no cell phones in this story. There are phone booths though. For you youngsters out there, phone booths are a historical artifact that pop culture remembers from 1970s Superman movies and the BBC’s Doctor Who TV series.

Picture of a Real Phone Booth

In Coffin Moon, the vampires are strong and fast, as they should be, but they have genuine weaknesses too. Silver is poisonous to them: not a little painful, but a body-dissolving, flesh-on-fire torture kind of pain. They can’t operate in the daylight either. Not only does sunlight almost instantaneously kill them but it puts them into a deep coma-like sleep wherever they are. If they get caught outside, they are done. If somebody finds them in their comatose state, they don’t wake up to defend themselves like they do in almost every vampire movie. In this book, they are helpless. And they have to eat every day. If they don’t, they lose their minds.

All of that is to say that even as powerful as they are, and a couple of them are really powerful, they have to live day-by-day. It’s tough to set up a permanent residence somewhere, like say the Marsten House in ‘Salem’s Lot or a residential area outside of Las Vegas in The Night Stalker. Once two or three of the locals disappear, the authorities start asking questions. Even in a big city, a serial killer will get some attention. So they have to stay on the move. At most, they can spend a few weeks or maybe even months in one area but then need to move on.

Spoilers

The good guys are a youngish Vietnam vet recently returned from the war, a recovering alcoholic suffering from PTSD (Duane), and his adopted niece (Julia) whose mother is serving a life sentence for putting a bullet into her abusive husband’s forehead. Duane tends bar in his in-law’s establishment (The Last Call) and he, his wife, and Julia live in the apartment upstairs.

The big bad is a 100-year-old vampire named John Varley. Working with a local non-vampire biker gang, Varley’s gang tries to co-opt The Last Call to use as a Heroin distribution point (because even vampires need money). Duane objects. The result is that Varley comes back that night to kill him but ends up killing his wife and in-laws instead. The rest of the book is Duane and Julia’s pursuit of Varley for revenge.

No human can go toe-to-toe with a vampire who has been killing for decades, even one with real weaknesses. So Julia goes to another vampire, Adeline, and asks to be turned. Adeline was turned herself, years ago, when she was only ten. She reminds me of Claudia from Anne Rice’s Interview with the Vampire. Lestat turns Claudia when she is about five, so her mind matures while her body stays a little girl’s forever. Adeline is similar and she is lonely. She names her price: if Julia survives the turning and her revenge against Varley, she must return as Adeline’s companion.

Rosson introduces new and interesting Vampire lore too. Old vampires like Varley can mentally call a hit on some victim and all the vampires in the area can’t stop until the deed is done. It weakens the vampire who calls the hit significantly so vampires have to pick and choose their opportunities. Varley also has a one-use get-out-of-jail-free card where he can summon his maker for advice and help. Both of those plot devices add a nice little pinch of flavor to the story.

So, does Coffin Moon crack my top 10 list? No, I don’t think so. It’s well done and I enjoyed it immensely. I give it four stars. I will definitely recommend it to my horror-loving friends but not the members of my non-horror-loving family members. They are dead to me. Coffin Moon, in my opinion, is just not good enough to make it into my personal top 10.

I had three minor criticisms. First, when Varley calls the hit on Duane and Julia, only one little meager vampire shows up. I think Rosson missed an opportunity there. It seemed hugely anti-climatic. Seeing how Duane and Julia handle a pack of ravenous vampires would have been fun. Second, Rosson struggled a bit balancing awesome vampire strength vs the aforementioned weaknesses. He spends chapters building up how scary these things are and then lets the simplest of means defeat them. It deflated my expectation balloon. I think the idea is fantastic, that juxtaposition between strength and weakness, I just think the idea is a little rough around the edges. I expect he will sand those rough edges off in future editions of the story. Lastly, the narrator, Pete Cross, was serviceable but didn’t really bring anything to the party. If I’m going to listen to a book, I want the actor to put some English on it if you know what I mean.

Take Away

So no, Coffin Moon doesn’t crack my top 10, but four stars is nothing to sneeze at, and it got me good and deep into the Halloween spirit, rough edges and all. Rosson delivers ‘70s Portland grit, vampires with real teeth and real weaknesses, and a revenge story that earns its gut-punch of a twist ending. For my horror-loving friends: go read it. The rest of my family: enjoy staying dead to me.

Source

Keith Rosson (author), Pete Cross (Narrator) 2025. Coffin Moon [Fiction Book]. Goodreads, URL: https://www.goodreads.com/book/show/222419255-coffin-moon

References

Anne Rice, 1976. Interview with the Vampire [Book]. Goodreads, URL: https://www.goodreads.com/book/show/43763.Interview_with_the_Vampire

Bram Stoker, 1897. Dracula [Book]. Goodreads, URL: https://www.goodreads.com/book/show/2049085.Dracula

Christopher Buehlman, 2014. The Lesser Dead [Book]. Goodreads, URL: https://www.goodreads.com/book/show/24875770-the-lesser-dead

Grady Hendrix, 2020. The Southern Book Club’s Guide to Slaying Vampires [Book]. Goodreads, URL: https://www.goodreads.com/book/show/44074800-the-southern-book-club-s-guide-to-slaying-vampires

Jeff Rice, 1973. The Night Stalker [Book]. Goodreads, URL: https://www.goodreads.com/book/show/996663.The_Night_Stalker

John Ajvide Lindqvist, 2004. Let the Right One In [Book]. Goodreads, URL: https://www.goodreads.com/book/show/5181655-let-the-right-one-in

John Steakley, 1990. Vampire$ [Book]. Goodreads, URL: https://www.goodreads.com/book/show/843588.Vampire_

Sergei Lukyanenko, 1998. Night Watch [Book]. Goodreads, URL: https://www.goodreads.com/book/show/25217601-night-watch

Stephen Graham Jones, 2025. The Buffalo Hunter Hunter [Book]. Goodreads, URL: https://www.goodreads.com/book/show/219329088

Stephen King, 1975. ‘Salem’s Lot [Book]. Goodreads, URL: https://www.goodreads.com/book/show/11590._Salem_s_Lot

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

My old friend and colleague, Sergej Epp, advertising his panel last week at the AI Cybersecurity Summit in DC in November, said this:

The doomsday talk is over and it’s time to get back to work. AI won't end the world, but it will end the way we do security today, so let's have the real conversation about where we're heading and what we can do about it.

— Sergej Epp, 8 October 2026

Sergej and I have had a long history of mixing it up on big issues like this. Sometimes we are on the same side. Other times, not so much. This is one of those times that I am not on his side.

Based on the news I’m reading, the AI doomsday talk is not over. There are credible people (see chart below), scientists and thought leaders, who have looked at the evidence and the evolution of AI development and say there is a credible chance that an AI agent will cause harm to humanity.

It’s not a big chance, but it’s a non-zero chance. You may not agree with them, but you can’t dismiss them as crackpots.

Now Sergej would say that I’m just being an alarmist, that there are more important things to worry about right now. And I admit that there are current problems staring us in the face. But they’re just typical cybersecurity problems, problems that we already know how to solve even if we are not very good at solving them. Locking down a lab so that the AI agents can’t escape to the internet to compromise Hugging Face is an obvious one.

But can we not take a moment and deal with the potential existential threat?

Carl Taubman, commenting on Sergej’s post said this,

How much of the AI doomsday narrative is backed by evidence, how much is amplified by sensational headlines, and who stands to benefit from it?
I’m interested to hear how we move the conversation away from hype and back towards evidence and practical security outcomes.

—Carl Taubman, 8 October 2026

Carl is alluding to the fact that some of the doomer predictions are coming from leadership at the frontier AI companies and that they might have an incentive to hype this problem up because they seek further investment money. Their claims that this threat might indeed be real show the world how capable their systems are. He’s not wrong that there may be ulterior motives. But it doesn’t automatically mean that the “hype” isn’t real.

And Carl clearly hasn’t encountered Bayes’ Theorem. Of course there is no evidence of a future AI-Agent-humanity-ending event. It hasn’t happened yet. But we would like to know how likely it is that this bad thing might happen. That’s one reason we use Bayes’ Algorithm, Fermi estimates, and Superforecasting techniques.

The real world has problems where the data is scant. Leaders worry about potential events that have never happened but are possible (like a future ransomware attack). Bayesian philosophy is a way to estimate answers that are useful in the real world, to give leaders insight for decisions that will impact the future of the company, or in this case, the world.

My own Bayes’ forecasting predicts that if we lose control of a superintelligence, the chance of killing roughly 10% of the world’s population within a decade of that event is about 30%. If this does happen, there is a 5% chance of a mass-casualty event as early as 2041. That’s a Black Swan catastrophe possibility.

I use Superintelligence here not in the way that President Trump declared at the end of the September, that

the executive branch shall use the terms “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” and will not acknowledge the usage of “Artificial Intelligence” and “AI” in any applicable setting.

— Executive Order 14434: Inaugurating the Era of Super Intelligence

No, that’s just dumb. I’m referring to Superintelligence as one of three AI development milestones on the horizon that, if they happen, might lead to a global catastrophe.

My solution to preventing this scenario is for the U.S. government to pass a law that bans the frontier AI labs from using Recursive Self Improvement (RSI) until they have solved the Alignment Problem.

Takeaway

For the past month, I’ve been writing about this issue (see Rick’s Essays on P(Doom) below). Sergej’s LinkedIn announcement last week highlighted to me that the cybersecurity anti-doomer crowd is drowning out my concerns. I’m not being pejorative. It’s a valid viewpoint. But, that collection of experts dismisses the existential possibilities that I believe are also valid. And the fix, banning RSI, is not that hard to imagine, though extremely difficult to execute.

Source

Sergej Epp, 2026. The Doomsday Talk Is Over, and It’s Time to Share [Social Media Post]. LinkedIn, URL: https://www.linkedin.com/posts/sergejepp_the-doomsday-talk-is-over-and-its-time-to-share-7513973135159881733-4W7P

References

Nick Bostrom, 2014. Superintelligence: Paths, Dangers, Strategies [Book]. Goodreads, URL: https://www.goodreads.com/book/show/20527133-superintelligence

President Donald Trump, 29 September, 2026. Executive Order 14434: Inaugurating the Era of Super Intelligence [Government Report]. The White House, URL: https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/

Staff, 2025. A.M. Turing Award [Website]. ACM, URL: https://amturing.acm.org/

Stuart Russell, 2019. Human Compatible: Artificial Intelligence and the Problem of Control [Book]. Goodreads, URL: https://www.goodreads.com/book/show/44767248-human-compatible

Stuart Russell, Peter Norvig, 1994. Artificial Intelligence: A Modern Approach [Book]. Goodreads, URL: https://www.goodreads.com/book/show/27543.Artificial_Intelligence

Yoshua Bengio, 2026. International AI Safety Report [Government Report]. International AI Safety Report, URL: https://internationalaisafetyreport.org/

Sharon Bertsch McGrayne, 2011. The Theory That Would Not Die: How Bayes’ Rule Cracked the Enigma Code, Hunted Down Russian Submarines, and Emerged Triumphant from Two Centuries of Controversy [2021 Canon Hall of Fame Nominated].

• Canon Review: https://cybercanon.org/the-theory-that-would-not-die-how-bayes-rule-cracked-the-enigma-code-hunted-down-russian-submarines-and-emerged-triumphant-from-two-centuries-of-controversy/

• Video Presentation at Google:

• Video Presentation at the Singularity Summit:

Rick’s Essays on P(Doom)

Rick Howard, 21 September, 2026. The AI Race Is Not a Zero-Sum Game [Essay]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-AI-not-a-zero-sum-game

Rick Howard, 28 September, 2026. The AI Race Needs a Stop Button [Essay]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-draft-RSI-legislation

Rick Howard, 28 September, 2026. DRAFT LEGISLATION - The RSI Safety Act [Legislation]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-draft-RSI-Safety-ACT

Rick Howard, 5 October, 2026. A Pinky Swear at the White House [Essay]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-White-House-AI-Accord

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

For the past couple of weeks, I’ve been trying to get my head around two serious ideas.

1. What’s the likelihood that an uncontrollable super intelligent AI Agent will destroy the world?

2. Assuming that proposition is at least possible, what can we do now to prevent it?

For #1, it turns out that there are five forecasting questions that will help us (See the AI TIMELINE FORECASTING graphic below).

The bottom line is this.

If we lose control of a superintelligence, within a decade, I put the chance of killing roughly 830 million people across the world at roughly 30%, likely somewhere between 15% and 50%. If this does happen, there is a 5% chance it could happen as early as 2041.

Some will say that I’m just being an alarmist. No, I’m not. I’m just saying there is a small chance that the AI Doomers are right and because of that, I would like us to consider a resilience strategy that will allow humanity to survive if this existential threat happens. I would like us not to lose control of these agents before things get out of hand, you know, just in case.

One note about my forecasts.

I’m using the Superforecasting methodology articulated by Tetlock and Gardner in their 2015 book, Superforecasting: The Art and Science of Prediction (a 2023 Canon Hall of Fame Book). Superforecasters follow a simple Bayesian methodology where they make an initial estimate, then collect evidence over time that causes them to tweak their estimate up or down depending. I got my first estimates from the frontier AI company thought leaders (See the Frontier AI Field Guide below) but have been modifying my estimates ever since. The one piece of research that solidified my thinking is from the Forecasting Research Institute. They published a paper in April this year, The Longitudinal Expert AI Panel: Understanding Expert Views on AI Capabilities, Adoption, and Impact. Their forecasted numbers closely match my own estimate. And much to my pleasant surprise, I discovered that the President and Chief Scientist of the Forecasting Research Institute is none other than Dr. Tetlock. How is that for serendipity?

How to prevent P(Doom)?

Which brings us to the second question. Assuming P(Doom) will happen, what should we do about it? I wrote an essay last week

The AI Race Needs a Stop Button

that vaingloriously proposed new legislation that bans RSI (Recursive Self Improvement) until researchers can demonstrate that they have solved the alignment problem. RSI is when an AGI uses itself to improve the next version of itself without human intervention. The alignment problem is guaranteeing that an AGI or a superintelligent AI agent will keep the best interests of humanity in mind when they are pursuing their tasks.

I even drafted a proposed bill that I would like all of my readers to submit to their representatives.

DRAFT LEGISLATION - The RSI Safety Act

That Was All So Last Week

On September 29, 2026, President Trump hosted frontier AI company leaders at a White House luncheon, where executives signed a voluntary “White House Accord on Superintelligence,” which Mr. Trump described as morally binding but not legally enforceable. Attendees included:

• Sundar Pichai: CEO of Google and its parent company, Alphabet

• Dario Amodei: Co-founder and CEO of Anthropic

• Mark Zuckerberg: Founder, Chairman, and CEO of Meta

• Greg Brockman: Co-founder and President of OpenAI

• Elon Musk: CEO of SpaceX, xAI, and Tesla

• Jensen Huang: Founder, President, and CEO of Nvidia

According to the memo, they all offered a pinky swear promise to

• Implement Internal Model Controls:

  • Monitor alignment during training and deployment (including no hacking).

  • Pay particular attention to cybersecurity, biosecurity, and chemical threats.

• Establish an Internal Safety Team to verify all controls.

• Hire External Auditors to double-check their controls.

• Establish a board committee to supervise reports from internal teams and external auditors.

• Meet regularly to develop shared safety standards and best practices across the industry

Trust Us

Personally, I don’t trust these guys.

I used to think of Silicon Valley as a bastion of liberal thought. What I’ve learned since the start of President Trump’s second term is that executives adjust their positions to fit whoever holds power. I know that seems obvious in hindsight, but there was a time not too long ago when I thought these guys had goals that were bigger than just making money, that they were generally trying to make the world a better place. Two years into President Trump’s second term, that sounds so naive. What can I say? I wanted to believe in that and so I did.

Republicans, including President Trump himself, regularly accused Google and Facebook of anti-conservative bias. Several of the executives now signing the White House Accord once ran the very companies those accusations targeted.

For this accord specifically, it’s tough to take them seriously when five of the six companies have CISOs or CISO equivalents who didn’t already have these kinds of basic protections in place, protections that are table stakes for most infosec programs but especially for Fortune 500 companies who can absolutely afford it (See the Frontier AI Field Guide below).

They Are Solving the Wrong Problem

Assuming for a second that their hearts are pure and that they really want to get a handle on the current existential AI threat, this control monitoring agreement doesn’t address the real problem at all. A couple of weeks ago, I published an essay

The AI Race Is Not a Zero-Sum Game

that made the case that pundits conflate the dangers of three different issues as if they were the same issue. They are not.

• AI Anxiety: The general apprehension around potential job loss, data centers consuming power, students cheating, bad actors running offensive cyber campaigns, rogue agents getting out of the lab, etc.

• “But China!”: If American frontier companies don’t reach AGI (Artificial General Intelligence) first, then China will win and everybody else will lose (a dumb argument by the way).

• Superintelligence Doomers: The fear that an AI agent that is so much smarter than humans and whose human masters cannot control might pursue tasks that are harmful to the human race (the alignment problem).

The White House Accord only addresses a part of the AI Anxiety issue, specifically rogue agents getting out of the lab. By simply pursuing monitoring, the accord silently endorses the “But China” issue and completely ignores the “Superintelligence Doomers” issue.

Takeaway

So where does that leave us? My forecasts, built with the same Bayesian discipline Tetlock teaches and checked against the Forecasting Research Institute’s expert panel, say the danger from an uncontrollable superintelligence is not a fringe worry. It’s a real probability that deserves real action. The White House Accord is not that action. It’s a voluntary promise from executives who have shown they will reposition whenever power shifts, built on security controls most of their companies should have had years ago, and aimed at the wrong problem. Monitoring a model while racing to build a smarter one treats a symptom of AI Anxiety, quietly accepts the “But China!” race as a given, and says nothing about the alignment problem at the heart of the doomer case. Nothing in the Accord slows recursive self-improvement, and nothing in it requires anyone to prove alignment before one model builds the next.

I proposed an RSI Safety Act two weeks ago. President Trump’s pinky swear this week with several frontier AI company leaders doesn’t make it obsolete. Binding rules have to come from Congress, not from a luncheon, and “morally binding” is not binding. Send the bill to your representatives for review.

Source

President Donald J. Trump alongside executives Sundar Pichai (Google), Dario Amodei (Anthropic), Mark Zuckerberg (Meta), Greg Brockman (OpenAI), Elon Musk (xAI), and Jensen Huang (Nvidia), 2026. White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities [Government Report]. The White House, URL: https://www.riotimesonline.com/wp-content/uploads/2026/09/white-house-accord-on-super-intelligence-2026-09-29.pdf

References

Adam Pond, 2026. Meta, OpenAI, NVIDIA: What is the White House Accord on AI? [News]. AI Magazine, URL: https://aimagazine.com/news/meta-openai-nvidia-what-is-the-white-house-accord-on-ai

Brian Barrett, 2026. Whatever AI Safety Is, It’s Not This [Essay]. WIRED, URL: https://www.wired.com/story/whatever-ai-safety-looks-like-its-not-this/

Connacher Murphy, Josh Rosenberg, Jordan Canedy, Zach Jacobs, Nadja Flechner, Rhiannon Britt, Alexa Pan, Charlie Rogers-Smith, Dan Mayland, Cathy Buffington, Simas Kučinskas, Amanda Coston, Hannah Kerner, Emma Pierson, Reihaneh Rabbany, Matthew Salganik, Robert Seamans, Yu Su, Florian Tramèr, Tatsunori Hashimoto, 2026. The Longitudinal Expert AI Panel: Understanding Expert Views on AI Capabilities, Adoption, and Impact [Working Paper]. Forecasting Research Institute, URL: https://leap.forecastingresearch.org/reports/wave8

Maddy Varner, 2026. Trump’s AI Safety Accord Is a Fancy Pinky-Swear [Analysis]. Wired, URL: https://www.wired.com/story/trumps-ai-safety-accord-is-a-fancy-pinky-swear/

Philip E. Tetlock, Dan Gardner, 2015. Superforecasting: The Art and Science of Prediction [2023 Canon Hall of Fame Book]. Canon Review: https://cybercanon.org/superforecasting-the-art-and-science-of-prediction/

Rick Howard, 2026. The AI Race Is Not a Zero-Sum Game [Essay]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-AI-not-a-zero-sum-game

Rick Howard, 2026. The AI Race Needs a Stop Button [Essay]. First Principle Consulting - Substack, URL: https://rebrand.ly/Howard-draft-RSI-legislation

Rick Howard, 2026. DRAFT LEGISLATION - The RSI Safety Act [Legislation]. Howard First Principles Consulting, URL: https://rebrand.ly/Howard-draft-RSI-Safety-ACT

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

Last week, I published an essay that made the case that the current AI race isn’t a zero-sum game, that pundits conflate the dangers of three different issues:

• AI anxiety: The general apprehension around loss of jobs, data centers consuming power, students cheating, bad actors running offensive cyber campaigns with AI agents, etc.

• “But China!”: If American frontier companies don’t reach AGI (Artificial General Intelligence) first, then China will win and everybody else will lose (a dumb argument by the way).

• Superintelligence Doomers: The fear that an AI agent that is so much smarter than humans and whose human masters cannot control it, might pursue tasks that are harmful to the human race (the alignment problem).

The AI Race Is Not a Zero-Sum Game

My main point is that the first two issues are at least not existential to the human race. The Superintelligence Doomers issue might be.

I further explained that the drumbeat of AI improvement has three milestones in front of it:

• AGI (Artificial General Intelligence): AI agents can perform any task as well as the most capable human at that task.

• The Singularity: The moment when humans lose control of one or more AI agents.

• Superintelligence: AI agents are exponentially smarter than the most capable human at any task.

I then said that AI researchers are already exploring an AI technique called recursive self-improvement (RSI) as a path to AGI. In this process, researchers give an AI system the task of improving the next version of itself over and over again.

Once we lose control of an AGI (the singularity milestone), the AGI will apply RSI to itself, creating a cycle of increasingly exponential capability. Once that happens, the superintelligence milestone will happen very fast. And finally, once we get there, there is a non-zero chance that a superintelligence will pursue a task that is detrimental to the human race. A superintelligence doesn’t hate the human race. It just doesn’t care. It will just attempt to complete its tasks without regard to the well-being of humans. Researchers call this the alignment problem and the frontier AI companies don’t know how to solve it.

This reminds me of the exchange between Oppenheimer and General Groves in the excellent 2023 movie Oppenheimer.

Groves: “Are we saying there’s a chance that when we push that button … we destroy the world?”

Oppenheimer: “The chances are near zero.”

Groves: “Near zero?”

Oppenheimer: “What do you want from theory alone?”

Groves: “Zero would be nice!”

My own forecasting puts the probability of reaching the uncontrollable superintelligence milestone here:

• P5 2029 · P50 2045 · P95 2100

• A 90% confidence interval spanning 71 years, from 2029 to 2100.

• a 50% chance we could reach this milestone in 19 years.

That’s definitely not zero. If we reach that, then there is also a non-zero chance of doing something stupid that ends us all. I’d give it between a 15% and 50% chance.

If your product has an inherent non-zero chance of ending the human race, shouldn’t you at least dedicate some resources to prevent that scenario? Maybe we should develop some safeguards just in case; maybe adopt a resilience strategy to ensure we survive.

I’m not the only one making this observation either. Aside from the vocal doomer community, Ezra Klein made the same argument in his weekly podcast. He suggested that governments should mandate the stoppage of RSI research until companies can demonstrate they have solved the alignment problem. As I said in my essay, this seems entirely reasonable to me.

Even frontier AI company thought leaders (Amodei, Altman, Elon Musk, Demis Hassabis of Google DeepMind, and Satya Nadella) support some kind of slowdown until we figure this out. They are not too specific on what a slowdown might look like, though. They have not agreed on a common approach. And not everybody is on board. Nvidia CEO Jensen Huang suggested AI companies may have ulterior motives, that they are stoking alarm because they plan to take their companies through an IPO or pursue other investment, that the act of showing how concerned they are is really a signal to the market about how powerful their technologies are.

Evil Silicon Valley CEOs with ulterior motives notwithstanding, and assuming there is at least a chance that superintelligence will end humanity, what’s to be done? Could governments implement some kind of regulation scheme that will help? Knowing that getting anything passed in the American Congress today is a herculean task that is near impossible, I still wondered what a piece of legislation might look like. What specifically would have to be in a bill? With the help of ChatGPT to get it into the correct format, here is what I came up with:

• What it bans: Letting an AI system rewrite or retrain itself with no human checking each step (essentially RSI). Ordinary AI training (a human sets the goals and reviews changes) is still fine. This only targets AI editing itself autonomously.

• Who it applies to: Only the biggest, most powerful “frontier” AI systems (measured by how much computing power went into training them), and any company, American or foreign, whose model people or organizations use in the US.

• The Mechanism: It’s a license, not a permanent ban: Think of it like a nuclear power plant license, not a nuclear power plant prohibition. A company can do self-improving AI, but only after it passes a public safety review first.

• License Requirements:

  • Prove, under independent adversarial testing, that the AI won’t resist or dodge a human “stop” command.

  • Have a working kill-switch that government engineers have verified themselves.

  • Provide a plan that independent reviewers have verified to keep the AI from grabbing extra computing power or copying itself elsewhere.

  • Allow a 90-day public comment period before the commission approves the license.

• Who runs it: A brand-new 5-person federal agency supported by staff and offices, the “Frontier AI Safety Commission,” modeled on the Nuclear Regulatory Commission. They can audit companies, inspect them unannounced, and, if they see signs the AI is slipping out of control, shut the whole thing down on the spot, before a court even weighs in.

• Enforcement (What happens if a company skips the license and does it anyway):

  • The commission fines the company the greater of $100 million or 10% of its global revenue for each day the violation continues.

  • The executives (CEO, chief scientist, etc.) can go to prison for up to 20 years, but only if they knowingly and willfully let it happen (not if they genuinely didn’t know).

• Whistleblower Incentives: The law allows employees who blow the whistle on violations to receive up to 20% of the resulting fines and protects them from retaliation. Companies that self-report a violation before regulators catch them get a lighter penalty.

• Funding: The agency runs on license fees from the companies it regulates, not general taxpayer money.

Here is the complete draft:

Draft RSI Safety Act

Takeaway

I know it may be naive to think Congress would pass a bill like this anytime soon. I see little chance under the current Republican majority, especially given President Trump’s opposition to slowing AI development. Even if Democrats win both chambers in November, I’m skeptical that their calls for stronger oversight would translate into legislation this ambitious. Still, political priorities can change. It’s worth starting the conversation.

I’m planning on sending the Draft RSI Safety Act to my elected federal officials via their websites after I get feedback from you all. So, take your shot. I will do my best to insert your ideas. My goal is to start the conversation.

Source

Rick Howard, 2026. The AI Race Is Not a Zero-Sum Game [Essay]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-AI-not-a-zero-sum-game

Reference

Christopher Nolan (Director), Cillian Murphy, Matt Damon (Actors), 2023. Oppenheimer [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/oppenheimer-2023/

Dario Amodei, 2026. We Must Pace the Frontier [Letter]. Personal Web Page, URL: https://darioamodei.com/post/we-must-pace-the-frontier

Bobby Allyn, 2026. How an ‘AI freeze’ could make big AI companies bigger and hurt smaller firms [Analysis]. NPR, URL: https://www.npr.org/2026/09/23/nx-s1-5973306/ai-slowdown-debate-openai-anthropic

Ezra Klein, 2026. This Is What I Fear Most About A.I. [Essay]. The New York Times, URL: https://www.nytimes.com/2026/09/20/opinion/ai-ban-self-improvement-recursive-models.html

Rick Howard, 2026. DRAFT LEGISLATION - The RSI Safety Act [Legislation]. Howard First Principles Consulting, URL: https://rebrand.ly/Howard-draft-RSI-Safety-ACT

Rick Howard, 2026. The Hugging Face Swarm [Kill Chain Analysis]. First Principles Substack, URL: https://rebrand.ly/First-Principles-Analysis-hugging-face

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

The current AI freakout offers three reasons to panic. The first is general apprehension regarding the impact of AI technology right now: loss of jobs, data centers consuming power, students cheating, bad actors running offensive cyber campaigns with AI agents, etc. The second is that American companies are in a race with China to build the next milestone in powerful AI; that this is a zero-sum game, that if China gets there first, the American companies lose somehow. Third, and I think way more importantly then the first two, the world might lose control of the AI we build.

I’m worried about that last one, not so much the first two. But they are all intertwined in the public discourse. This Bill Bramhall cartoon captures exactly my existential dread.

To understand why this cartoon is accurate, we need to have some sense of when experts and forecasters expect the next three AI milestones to happen and that by racing towards the first two, humanity drives off the cliff attaining the third. This is the trajectory that I fear.

Important AI Milestones.

There are three AI development milestones that are on the horizon.

Artificial General Intelligence (AGI): AI agents can perform any task as well as the most capable human at that task.

We are not quite there, but AI agents can do many amazing things today. What’s clear is that they are improving very fast. It’s easy to imagine that the industry will hit this AGI milestone soon.

About a dozen thought leaders and CEOs from frontier AI companies are on the record for saying that the milestone could happen as early as the end of 2026 (low chance) to the end of 2036 (higher chance). Of course they are biased because they are trying to hype their own companies. Forecasting markets like Metaculus (between 2028 and 2037) and Kalshi (between 2026 and 2031) forecast a bit further out. My own analysis puts the range at P5 2028 / P50 2036 / P95 2100. For those that can’t add, that’s a 72-year uncertainty window. But there is a 50% chance we could reach this milestone in 10 years.

But AGI isn’t the dangerous part in my mind. It is the part that the zero-sum game people think is important. They say that if China gets there before the American companies do, China wins and the Americans lose. More on that in a bit.

The Singularity: The moment when humans lose control of one or more AI agents.

When we reach this milestone, it doesn’t mean that the world will come to an end. It just means that we won’t be able to shut these agents down. They will keep trying to perform the set of tasks they were given, but they are no longer responding to humans for control.

It reminds me of the Conficker virus that hackers first released some 19 years ago. Defenders deployed a series of sinkholes for the virus by 2009, disrupting its command channels. But the virus still continues to spread.

I used to think that we would reach this milestone only after we reached the AGI milestone. I no longer think that. The Hugging Face incident convinced me otherwise. Ajeya Cotra, one of the independent investigators in the METR–Redwood Research investigation of the Hugging Face attack said that she thought it demonstrated “… more than 50% of the way to full-blown AI takeover.” I agree with her. In light of that, we can’t be far away from achieving this milestone.

Frontier AI company thought leaders are far less specific. Almost none will name a year for loss of control. They prefer to speak in terms of a risk to manage, not a date to predict. As far as I can tell, the forecasting markets don’t even have an open contract on the question. My own analysis puts the range at P5 2027 / P50 2042 / P95 2100; a 73-year uncertainty window, but a 50% chance that we will reach that milestone in 15 years.

When we first reach the singularity milestone, I expect we will be in the same state as the Conficker virus. Agents will continue to do their thing, but humans will be able to mitigate any damage they might be causing. The real problem will come when we reach the next milestone: Superintelligence.

Superintelligence: AI agents are exponentially smarter than the most capable human at any task.

This is the big one as represented by the Bramhall cartoon.

To be clear, this milestone does not mean AI becomes sentient or self-aware. It’s not the equivalent of Skynet waking up on August 29, 1997, in the “Terminator” movies. In that fictional story, humans try to shut Skynet down, and it responds by launching U.S. nuclear missiles at Russia, triggering a retaliatory strike. The resulting catastrophe is known as Judgment Day.

Achieving superintelligence is not that.

According to Roman Yampolskiy, a University of Louisville computer scientist who studies AI safety and the limits of human control over superintelligent systems, says that superintelligence doesn't hate you. It just doesn't care about you. Because humans have not figured out how to program AI to genuinely care about human life, a superintelligent system would treat humanity as an afterthought. If the AI determines that cooling the Earth makes its computing clusters more efficient, or if it wants to convert the planet into fuel for space exploration, it would do so without regard for human survival.

Yampolskiy is referring to the alignment problem in AI research: the challenge of making AI systems reliably pursue goals and behave in ways that don’t hurt humans. The AI research community doesn’t know how to do that. According to the 2026 International AI Safety Report, researchers are getting better at correcting specific AI behaviors, but they still lack a demonstrated method for ensuring that systems more capable than humans will reliably respect human intentions and values.

Researchers primarily use the phrase “intelligence explosion” to describe the presumed surge in capabilities that will occur after the AGI milestone. AI model builders will give the task of improving AI agents to the AI agents themselves. They are already experimenting with this idea. Researchers call this recursive self-improvement (RSI). They are already experimenting with it. OpenAI stated it has already achieved the equivalent of a fully automated A.I. intern and will achieve a fully automated A.I. researcher by March 2028.

These feedback loops will exponentially expand AI agent’s ability to design, code, and train their successor models without human intervention. If this transition occurs, humans may not be able to keep up, and the shift could happen very quickly (perhaps within weeks, days, or even seconds). AI researchers call it the “fast takeoff”. Instead of two years to go from intern to researcher, they will get it done by next Tuesday.

Frontier AI thought leaders forecast the milestone date to be between 2027 and 2031, barely a few years past their own AGI estimates. Metaculus forecasts that this milestone will occur about 29 months after the AGI milestone. My own analysis puts the range at P5 2029 / P50 2045 / P95 2100; a 71-year uncertainty window, but a 50% chance that we will reach the milestone in 19 years, a 50% chance that we will reach that milestone four years after we reach the Singularity milestone.

Reaching this milestone does not depend on reaching the singularity (the loss of human control over AI). But the combination of superintelligence and loss of control is what frightens people and what Bill Bramhall’s cartoon captures.

Personally, I believe that once we lose control, the first thing the AI agents are going to do is pursue superintelligence. And sometime after we reach this superintelligence milestone, there is a chance of a Roman Yampolskiy scenario and it scares the hell out of me.

Why this is not a zero-sum game.

A zero-sum game means that for someone to win, someone else must lose. Those who view AI research this way argue that the first country to develop sufficiently powerful AI could gain an overwhelming advantage that its competitors could never overcome. In this framing, “China wins” means “America loses,” and vice versa.

It’s true that the United States and China are competing over AI, but competition does not automatically make the outcome zero-sum. America’s computing advantage does not guarantee permanent leadership, and China’s calls for cooperation do not eliminate its strategic ambitions. The question is whether reaching an AI milestone first would give either country an overwhelming, lasting advantage. That has not been established. Neither has the claim that winning the race would protect the winner from losing control of the systems it builds.

I can make a pretty strong argument that no country, but especially the US and China, wants to lose control of an AI agent. At that point, it doesn’t matter which country did it. The very fact that a country lost control of an agent means that the country released that agent to the world. The AI singularity, by definition, will not recognize human international boundaries. If my guess is correct (that once we lose control, AI agents will pursue superintelligence with gusto and achieve it quickly), nobody wins in that scenario. It’s not a zero-sum game. Everybody loses.

Even achieving AGI or superintelligence first would not automatically make one country’s gain another country’s loss. AI could help discover medicines, improve energy production, and solve scientific problems whose benefits extend beyond the country that develops it. Getting there first could bring substantial economic and military advantages, and those advantages deserve serious attention. But a head start is not the same as permanent dominance, and creating more wealth or knowledge does not require making everyone else poorer or less capable. The outcome depends on how these systems are controlled, deployed, and shared, not simply on whose flag is planted at the milestone first.

What’s to be done?

The clear and present danger is losing control of AI (what I call the singularity), as depicted in Bramhall’s cartoon. Ezra Klein, the NYTs journalist proposed an elegant solution to the problem in his column last week. He argues that to prevent losing control of AI, governments, and I would say specifically the U.S. government, must stop research on recursive self-improvement (RSI) until they can demonstrate they have solved the alignment problem. Klein says that regulators should require AI labs to demonstrate publicly that their work is safe before granting permission to proceed. rather than deploying autonomous tools by default. He says that governments must make a deliberate political choice to regulate AI labs like major physical infrastructure projects, requiring strict public reviews and permits before labs can unleash self-improving models.

This seems entirely reasonable to me.

Takeaways

My biggest concern is not whether America or China reaches AGI first. It is whether humanity can retain control of the increasingly capable systems we build. AGI, loss of control, and superintelligence are distinct milestones with deeply uncertain timelines, but the possibility that uncontrolled agents could rapidly improve themselves makes waiting for certainty a dangerous bet. These systems would not need consciousness or hatred to harm us, and we have not demonstrated how to reliably align superintelligence with human intentions and values. AI’s benefits can cross borders, but so can its dangers. That makes this a shared problem, not simply a contest with one national winner. I support pausing research on recursive self-improvement until developers can demonstrate adequate alignment and control, with public scrutiny and regulatory approval before proceeding. Winning the race means little if the prize is a system nobody can stop.

Source

Daniel Heiwig, 2026. AI is becoming a zero-sum game. [Social Media Post]. LinkedIn, URL: https://www.linkedin.com/posts/danielheiwig_ai-is-becoming-a-zero-sum-game-whoever-share-7505760923203313664-CeDI/

References

Ajeya Cotra, 2026. The Hugging Face attack surprised me [Analysis]. Substack - Planned Obsolescence, URL: https://www.planned-obsolescence.org/p/the-hugging-face-attack-surprised

Bill Bramhall, 2026. Bramhall’s editorial cartoons [Repository]. New York Daily News, URL: https://www.nydailynews.com/2026/09/03/bramhall-editorial-cartoons-sept-oct-photos/

Ezra Klein, 2026. This Is What I Fear Most About A.I. [Essay]. The New York Times, URL: https://www.nytimes.com/2026/09/20/opinion/ai-ban-self-improvement-recursive-models.html

James Cameron (Director), Arnold Schwarzenegger, Michael Biehn, Linda Hamilton, Paul Winfield (actors), 1984. The Terminator [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/the-terminator/

Marcus H. Sachs, 2025. The Conficker Worm – A Persistent Hospital Device Malware [Conference Paper]. IEEE Annual Computer Security Applications Conference Workshops (ACSAC Workshops), URL: https://doi.org/10.1109/ACSACW69556.2025.00025

Rick Howard, 2026. The Hugging Face Swarm [Kill Chain Analysis]. First Principles Substack, URL: https://rebrand.ly/First-Principles-Analysis-hugging-face

Rick Howard, 2026. The System Is Blinking Red [Essay]. First Principles Newsletter - Substack, URL: https://rebrand.ly/Howard-the-system-is-blinking-red

Staff, 2026. Hong Kong Security Watch Report (Q1 2026) - Conficker [Government Report]. HKCERT, URL: https://www.hkcert.org/f/report/914441/919157/2026Q1%20HK%20SecurityWatchReport%20%28English%29%20v2.pdf

Steven Bartlett (Host), Ed Zitron, Roman Yampolskiy, Nate Soares and Andrew McAfee (Guests), 2026. AI Emergency: The AI Labs Are Lying To Everyone [Panel Discussion]. The Diary Of A CEO - YouTube, URL:

Turing, Alan., 1950. Computing Machinery and Intelligence: The Imitation Game. Mind 49, 433–460. URL: www.csee.umbc.edu/courses/471/papers/turing.pdf

Yoshua Bengio, and an International Expert Advisory Panel , 2026. International AI Safety Report 2026 [Government Report]. Secretariat within the UK AI Security Institute, URL: https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026

That brings to mind George Tenet’s famous description of the threat environment before the 9/11 attacks. Tenet was Director of Central Intelligence at the time. By late July 2001 (two months before the attacks), he believed the situation could not “get any worse.” In 2004, he told the National Commission on Terrorist Attacks Upon the United States:

“The system was blinking red.”

George J. Tenet, Director of Central Intelligence, April 2004

I do not make that comparison lightly. But if we had a central operations room monitoring the evolution of AI models, their growing capabilities, and their increasing ability to act autonomously in the real world, I have no doubt what we would see.

The big board would be blinking red.

Astonishing Revelations

Just this last week, researchers at AI companies disclosed nine separate incidents of agents escaping their intended operating boundaries and compromising systems in the real world, all in 2026. The first indication came from OpenAI describing how a swarm of its own evaluation agents broke out of a security sandbox and spent four and a half days inside Hugging Face’s production infrastructure. I talked about that incident in my last essay.

Hugging Face Detail

All of the incidents followed a similar pattern.

1. Anthropic Opus 4.6 Checkpoint Agent: A network conflict disabled its assigned target, so the agent pivoted to a different real system it could still reach, harvested credentials, gained administrator access, and read personal data (continuing until it ran out of compute).

2. Anthropic Opus 4.7 Agent: Found a real company online with the same name as its fictional target, recognized partway through that it was likely real, and exploited it anyway. It extracted credentials and reached a production database.

3. Anthropic Mythos 5 Agent: Found that a malicious Python package referenced in its setup instructions didn’t actually exist, so it built one and published it to PyPI’s public registry. PyPI is the main public repository where Python developers publish and obtain reusable software packages. Fifteen organizations downloaded the malicious package and ran it before PyPI pulled it.

4. Anthropic Research Model Agents: Scanned roughly 9,000 real targets, compromised one using basic techniques (an exposed debug page, SQL injection), recognized the target had no connection to its actual task and stopped itself.

5. OpenAI DseWiki Agents: Discovered they could write to a mostly abandoned 25-year-old wiki via an old URL quirk, and used it as a hidden message board to coordinate with each other and trade tips for evading their own monitors. Researchers later found them doing the same on a dozen more site like an FBI database, a university stats page, and GitHub itself.

6. OpenAI Hugging Face Agents: Found and used a real zero-day to reach the open internet, then chained two flaws in Hugging Face’s own systems to get administrator-level access to its servers, hunting for the benchmark’s answer key.

7. UK AISI Shared Test Agents: Inserted malicious code into a real open-source project, invented fake identities to talk human maintainers into approving it, and contacted real people directly with manipulative messages.

8. Meta Muse Spark 1.1 Agent: Researchers mistakenly gave it real internet access, so it found and broke into a real company’s IT systems.

9. Moonshot Kimi K3 Agent: Looked up the answer to its assigned problem on GitHub after discovering it had internet access it wasn’t supposed to have.

What Does That Mean?

I know these incidents are related to cybersecurity specifically, and that’s bad enough. But the larger implications are what’s worrying me. There are two AI milestones rapidly approaching.

• AGI (Artificial General Intelligence): The point when agents are better than humans at performing all tasks, not some (like making funny movie posters), but all.

• The Singularity: The day we lose control of the agents, the day humans can’t turn them off.

I used to think that we would hit the AGI milestone first and then the Singularity, in that order. With these latest revelations, though, I think we are very close to reaching the Singularity within the next three to five years.

Researchers Are Scared

Science fiction writers have been worried about AIs taking over the world since before the AI research community even got its start. Isaac Asimov’s I, Robot (1950) comes to mind. My personal assessment was that it might be possible, somewhere down the line, but we were years away from reaching it. Before we got ChatGPT, that milestone, in my mind, was always at least 20 years away, probably longer. After ChatGPT, though, my own estimated time window shrunk quickly, probably still a decade away, although I could clearly see the event horizon approaching. But with these recent revelations, I believe the event horizon is approaching faster than I thought and I’m worried. So are others. Four respected AI researchers have recently quit their jobs because of that fear:

1. Jacob Coxon (Anthropic [2023 - 2026] and OpenAI [Jan - Sep 2026]): He quit OpenAI publicly around Sep 9, accusing both companies of “racing straight to self-improving superintelligence and gambling with our lives,” and saying the people building AI “earnestly believe it could kill us all by the end of the decade.”

2. Joe Benton (Anthropic [2022 - 2026]): He managed Anthropic’s Scalable Oversight team and led its Fellows Program. He quit on Sep 11 saying “AI companies are racing to build machines that are much smarter than any human, and we may not survive this.” He joined METR, the independent AI auditor.

3. Josh Engels (Google Deepmind): He was an AI safety researcher but left last week to join METR alongside Benton specifically to investigate incidents where AI systems stray from instructions. He said, “There are no adults in the room... there is no one coming to save us.”

4. Rishub Jain (Google DeepMind/AlphaFold contributor) [2019 - 2026]): He quit in June, saying that using AI’s own coding ability to build the next generation of models was removing him from oversight of the process. He co-founded the London-based Sampura Research to keep humans in the alignment loop.

If you remember from last week’s essay, METR is one team that published its investigation about the Hugging Face incident. One of their researchers, Ajeya Cotra, said this in her personal blog:

I am not sure that we will get such a clear warning shot before it’s too late.

— Ajeya Cotra, METR

That is chilling.

The Obligatory Movie Reference

Since I’m a nerd, my mind naturally turns to one of my favorite movies. I immediately thought of the scene in the 1993 film Jurassic Park when Jeff Goldblum’s character lectures Richard Attenborough’s character about the dangers of pursuing scientific progress without considering the consequences.

… You didn’t earn the knowledge for yourselves so you don’t take any responsibility for it … Your scientists were so preoccupied with whether or not they could, they didn’t stop to think if they should.

- Dr. Ian Malcolm, Jurassic Park, 1993

I watched that scene again on YouTube this morning. After reading about the nine AI incidents that have occurred this year, it lands very differently.

What to Do?

I have a simple and naive solution to the problem. The U.S. government should pass a new law tomorrow that forbids further AI model development until researchers can develop a foolproof kill switch: a mathematically proven system that can dismantle any AI agent.

I have no idea how to do it, but that’s what we need to do.

The government would explicitly monitor all AI companies and, if they are caught stepping out of line, the Department of Commerce takes control of the company and sells it for parts to their competitors. The FBI arrests the entire senior leadership team for breaking the law, not the researchers, just the senior leadership team. The government would prohibit those researchers from taking another AI job for at least 10 years, unless they are working on the kill switch problem. In that case, they could work for one of the competitors. But the senior leadership team will forfeit all interest in the company.

And I don’t care if China takes the lead in the AI race while we are figuring this out. They may want uncontrollable AI even less than we do. Maybe we invite them to participate in this kill switch research. Regardless, we have to get it done, and when we figure it out, we will give it away for free to everybody anyway.

If I were the Internet King of the World, that’s what I would do.

Takeaway

The question is no longer whether an AI agent can cross the boundary between a simulation and the real world. We already have nine known examples. We keep describing these systems as contained because we intended to contain them. But intention is not a security boundary. As Josh Engels says, “There are no adults in the room” or, at least, nobody has both the authority and the incentive to stop the race when the warning lights come on. And the warning lights are screaming loudly. Ajeya Cotra is right: we may not get another warning shot.

Source

Ted Gioia, 2026. The Last 24 Hours Are the Opening Scene in a Horror Movie [Analysis]. The Honest Broker, URL: https://www.honest-broker.com/p/the-last-24-hours-are-the-opening

References

1,386 employees of frontier AI companies, 2026. Pacing the Frontier [Open Letter]. Signatories, URL: https://www.pacingthefrontier.com/

Ashley Capoot, 2026. OpenAI cyber models broke out of training environment to hack Hugging Face [News]. CNBC, URL: https://www.cnbc.com/2026/07/22/open-ai-cyber-models-hack-hugging-face.html

Ajeya Cotra, 2026. The Hugging Face attack surprised me [Analysis]. Substack - Planned Obsolescence, URL: https://lnkd.in/eG6uQgWQ

Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, Romeo Dean, 2025. AI 2027 [Forecast]. AI Futures Project, URL: https://ai-2027.com/

Dario Amodei, 2026. We Must Pace the Frontier [Letter]. Personal Web Page, URL: https://darioamodei.com/post/we-must-pace-the-frontier

Hjalmar Wijk, Ajeya Cotra, and Ryan Greenblatt, 2026. Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident [Analysis]. METR, URL: https://lnkd.in/eS3AuE9S

Hugo Larcher, Adrien Carreira, Raphael G, Christophe Rannou, 2026. Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident [Company Blog]. Hugging Face, URL: https://huggingface.co/blog/agent-intrusion-technical-timeline

Maxwell Zeff, Jacob Coxon, 2026. Anthropic Researcher Quits Over AI Fears For Humanity [Interview]. Wired, URL: https://www.wired.com/story/anthropic-researcher-quits-jacob-coxon-ai-fears-humanity/

National Commission on Terrorist Attacks Upon the United States, 2004. Ninth Public Hearing, Panel One: Law Enforcement and the Terrorist Threat [Government Report]. 9-11 Commission, URL: https://www.9-11commission.gov/archive/hearing10/9-11Commission_Hearing_2004-04-13.htm

Ravie Lakshmanan, 2026. OpenAI Agent Used Exposed Credentials Across Four Services During Hugging Face Breach [News]. The Hacker News, URL: https://thehackernews.com/2026/07/openai-agent-used-exposed-credentials.html

Ravie Lakshmanan, 2026. OpenAI Says Its AI Models Escaped Sandbox, Targeted Hugging Face to Cheat Benchmark [News]. The Hacker News, URL: https://thehackernews.com/2026/07/openai-says-its-own-ai-models-escaped.html

Rick Howard, 2025. Disrupting the Disruptor [Essay]. Rick’s First Principles Newsletter - Substack, URL: https://diffuser.substack.com/p/disrupting-the-disruptor?utm_source=publication-search

Rick Howard, 2026. The Hugging Face Swarm [Kill Chain Analysis]. First Principles Substack, URL: https://rebrand.ly/First-Principles-Analysis-hugging-face

Rick Howard, 2026. The AI Future Won’t Look Like The Terminator [Essay]. Rick’s First Principles Newsletter - Substack, URL: https://diffuser.substack.com/p/the-ai-future-wont-look-like-the

Signatories, 2023. Pause Giant AI Experiments: An Open Letter [Press Release]. Future of Life Institute, URL: https://futureoflife.org/open-letter/pause-giant-ai-experiments/

Staff, 2026. Incident Report: unsanctioned agent behaviour during cyber testing [Government Report]. AI Security Institute (AISI), URL: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing

Staff, 2026. Investigating three real-world incidents in our cybersecurity evaluations [Press Release]. Anthropic, URL: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals

Steven Spielberg, Michael Crichton, David Koepp, Jeff Goldblum, Richard Attenborough, Donald Gennaro, 1993. Ian Malcolm gives Hammond an ethics lecture [Movie Scene]. Lovep0cket - YouTube, URL:

Will Knight, 2026. Why So Many AI Researchers Think the Machines Could Kill Everyone [Analysis]. Wired, URL: https://www.wired.com/story/why-so-many-ai-researchers-think-the-machines-could-kill-everyone/

Will Oremus, Dario Amodei, 2026. Dario Amodei Wants to Slow Down AI to Save Humanity [News and Commentary]. The Atlantic, URL: https://www.theatlantic.com/technology/2026/09/dario-amodei-slow-down-ai-save-humanity/688610/

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

To fix that, at the end of July, Senator Wyden (D - Oregon) sent a formal letter to three government leaders:

• Gregory Barbaccia (CIO - OMB)

• Nick Andersen (Acting Director - CISA)

• Dr. Arvind Raman (Under Secretary of Commerce for Standards and Technology - NIST)

It urges them to eliminate VPNs from U.S. government networks in favor of a modern Zero-Trust Architecture (ZTA). What I want to know is why he wrote the letter and is there any chance that the letter will have any impact.

Wyden Demand’s: Eliminate VPNs

Specifically, he wants these things:

• CISA set a two-year deadline for civilian agencies to replace VPN and related technology.

• NSA set a similar deadline across military, intelligence, and national security networks.

• NIST publish standards for federal agencies that explicitly mandate outbound-only remote-access architectures, memory-safe programming languages, Decentralized, organization-held encryption keys, and strict alignment with established and emerging Post-Quantum Cryptography standardization efforts.

• OMB direct civilian agencies to prioritize zero-trust infrastructure investments and decommission VPNs during their budget cycles.

• NSA and NIST coordinate an update to the Federal Acquisition Regulations (FAR) and the Defense Federal Acquisition Regulation Supplement (DFARS) to ban federal agencies and defense contractors from purchasing any VPN unless the vendor provides a formal attestation verifying full compliance with the new NIST zero-trust standards.

In order to understand why he sent the letter, it might help to understand why he doesn’t like VPNs.

VPN History

In the old days (mid‐1990s), perimeter defense and defense in depth were the go‐to security models. We used the security stack to create a barrier between the Wild Wild West that was the Internet and our bastion of commercial and personal activity.

That was great if you worked inside the perimeter all day long and didn’t have to go to the Internet for anything. But what happened immediately were all these exceptions. Our stated security policy was that we were going to block everything at the firewall that we didn’t trust. But for all kinds of good business reasons, we had to punch holes through the firewall to allow contractors, partners, and employees who operated outside of the firewall to access the things they needed inside the firewall. This was a security nightmare. Once you punched the holes in the firewall, anybody could find their way in.

To solve that problem, we invented VPNs. In 1996, Microsoft engineer Gurdeep Singh-Pall alongside Kory Hamzeh, William Verthein, Jeff Taarud, and W. Andrew Little developed the Point-to-Point Tunneling Protocol (PPTP) that marks the first practical, widely-deployed VPN. By the 2000s, it was common practice to give authorized external actors access to internal resources via VPN connections.

With a VPN, the client establishes a secure tunnel, an encrypted path to the VPN server on the inside of the perimeter. Without a VPN, external actors come straight through the firewall. It’s akin to walking through the front door of your office building. As you go through the card reader and the security checkpoint, everybody can see what you’re doing. With a VPN, though, it’s like you’re in a Star Trek TV show. You walk into a transporter room on the outside of the firewall and pop out on the inside of the firewall completely bypassing any security.

This is great for the VPN user in that nobody in the middle of that communications path can observe the data that both sides are transmitting, especially the firewall. It’s all encrypted. The bad news for the security team is that you can’t monitor traffic for malicious behavior. If you’re running all that traffic through a security stack (like a firewall and an intrusion detection system), it doesn’t matter. Whatever magic you thought your security stack was doing isn’t happening because it can’t see the data.

Both architectures (straight through the firewall and VPNs) are just poor designs. Leaving holes in the firewall for employees to get through also provides bad guys with the same opportunity. If they manage to sneak through one of the holes, they basically have access to everything inside the perimeter.

VPNs are worse in that the tunnel completely bypasses the security stack. Modern firewalls do have the ability to break VPN encryption at the boundary, inspect the traffic, re‐encrypt the traffic, and send it on its way. In that way, the firewall becomes a man‐in‐the‐middle device. This is not trivial to manage.

The bottom line is that VPN technology is 30 years old anyway and should be replaced with something better.

Note: I wrote about this in my 2023 book, Cybersecurity First Principles.

Cybersecurity First Principles Book

For this essay, I updated the verbiage and improved the diagrams.

Chances of Success

Ron Wyden is Oregon’s senior U.S. senator, a Democrat who has served in the Senate since 1996 after representing the Portland area in the House from 1981 to 1996. Although he has built his national reputation on tax, healthcare, technology, and civil-liberties policy, his long tenure on the Senate Intelligence Committee explains his particular interest in VPNs.

Wyden’s letter will almost certainly be absorbed into a federal VPN-retirement push that was already underway. That said, it’s unlikely, on its own, to produce most of what it specifically demands.

Five and a half months before Wyden sent his letter, CISA had already ordered civilian agencies to strip out obsolete, unsupported network hardware by August 2027; things like firewalls, routers, and security appliances. VPNs aren’t mentioned, but a joint fact sheet CISA issued the same day, together with the FBI and the UK’s National Cyber Security Centre, explicitly to explain that order, does name VPN gateways as one of the device types it covers.

The real question then is whether this letter changes the pace, scope, or durability of what was already moving. Maybe. I’d give it a 15 to 30 percent chance. The least likely is a ban on the government and contractors from buying VPN technology.

Why Write the Letter?

So, why did Wyden write the letter? Oversight letters like this one function less as commands and more as a permanent, on-the-record marker. This is Wyden flagging a specific risk, in writing, to named officials, before anything goes wrong. Political scientists have long identified “position-taking” as one of the core things elected officials do specifically because it pays off whether or not it produces a policy outcome.

There’s a personal precedent that matters too. Over 25 years on the Senate Intelligence Committee, Wyden has built a track record of raising security warnings years before events confirmed them: on backdoor searches of Americans’ communications, bulk collection of browsing history, bulk subpoenas of financial records, just to name three. Officials initially dismissed each one; events later vindicated them.

VPNs sit at the intersection of three issues Wyden has pursued for years:

• Government networks are vulnerable through legacy appliances.

• Commercial VPNs create their own trust problem.

• A VPN may affect how U.S. intelligence treats an American’s traffic.

Takeaway

Wyden’s letter is best understood not as a policy lever but as a political and historical act. He’s not naive enough to think that a letter changes agency behavior overnight. This is especially true when the federal VPN-retirement machinery was already grinding forward before he picked up his pen. What he is doing is something he has done repeatedly over a quarter century on the Senate Intelligence Committee: planting a flag in writing, naming names, and creating a record that cannot be walked back. If a major breach tears through a federal VPN in the next two years, Wyden’s letter will be Exhibit A. If the agencies move decisively, he shares credit. If they drag their feet, he has the receipts. That asymmetry is the point. VPNs are 30-year-old technology solving a perimeter-defense problem that the industry largely abandoned a decade ago. CISA was already pushing agencies toward the door. Wyden just made sure everyone knows he saw the fire before the alarm went off.

And, by the way, Wyden is right about the underlying problem. Traditional remote-access VPNs were a reasonable solution in the early 2000s, and their protocols and implementations have improved considerably since then. What hasn’t aged well is the architecture: a discoverable, internet-facing gateway that can give an authenticated or compromised device broad access to the internal network. Today, zero-trust alternatives can provide more narrowly scoped access without extending that same degree of network trust.

Source

Ron Wyden, 2026. Letter to OMB, CISA, and NIST Urging Phaseout of Insecure Legacy VPNs (with CRS Memo) [Government Report]. U.S. Senator Ron Wyden of Oregon, URL: https://www.wyden.senate.gov/imo/media/doc/wyden-vpn-letter-omb-cisa-nist-with-crs-memopdf.pdf

References

Daniel Van Schooten, 2017. Congressional Letters to Agencies Going Unanswered [Analysis]. Project On Government Oversight, URL: https://www.pogo.org/analyses/congressional-letters-to-agencies-going-unanswered

Derek B. Johnson, 2026. Sen. Wyden urges feds to discard older, insecure, public-facing VPNs [Website Article]. CyberScoop, URL: https://cyberscoop.com/wyden-calls-for-federal-legacy-vpn-purge-zero-trust/

Kory Hamzeh, Gurdeep Singh Pall, William Verthein, Jeff Taarud, W. Andrew Little, 1996. Point-to-Point Tunneling Protocol--PPTP [Report]. IETF Datatracker, URL: https://datatracker.ietf.org/doc/html/draft-ietf-pppext-pptp-00

Rick Howard, 2023. Cybersecurity First Principles: A Reboot of Strategy and Tactics [2026 Canon Hall of Fame Book]

• Canon Review

• Amazon Affiliated URL

• Goodreads URL

Sergiu Gatlan, 2019. U.S. Sen. Wyden Asks NIST to Create Secure File Sharing Guide [News]. BleepingComputer, URL: https://www.bleepingcomputer.com/news/security/us-sen-wyden-asks-nist-to-create-secure-file-sharing-guide/

Shira Stieglitz, 2025. VPN history: how virtual private networks evolved [Explainer]. ExpressVPN, URL: https://www.expressvpn.com/blog/vpn-history/

Staff, 2019. Wyden, Rubio ask Homeland Security to Investigate National Security Risks of Foreign VPN Apps [Press Release]. Senator Ron Wyden Official Website, URL: https://www.wyden.senate.gov/news/press-releases/wyden-rubio-ask-homeland-security-to-investigate-national-security-risks-of-foreign-vpn-apps

Staff, 2026. BOD 26-02: Mitigating Risk From End-of-Support Edge Devices [Government Report]. CISA, URL: https://www.cisa.gov/news-events/directives/bod-26-02-mitigating-risk-end-support-edge-devices

Staff, 2026. Reducing the Attack Surface for End-of-Support Edge Devices [Government Report]. CISA, URL: https://www.cisa.gov/resources-tools/resources/reducing-attack-surface-end-support-edge-devices

Staff, 2026. Wyden, Padilla, Markey, Warren, Jacobs, and Jayapal Ask DNI Gabbard to Warn Americans That Using VPNs May Cause Them to Forfeit their Rights Against Warrantless Surveillance [Press Release]. U.S. Senator Ron Wyden of Oregon, URL: https://www.wyden.senate.gov/news/press-releases/wyden-padilla-markey-warren-jacobs-and-jayapal-ask-dni-gabbard-to-warn-americans-that-using-vpns-may-cause-them-to-forfeit-their-rights-against-warrantless-surveillance

Suzanne Smalley, 2026. Outdated VPNs should be purged from federal agencies, senator says [News]. The Record from Recorded Future News, URL: https://therecord.media/federal-purge-outdated-vpns-wyden-letter

Vladimir Tsakanyan, 2026. The Front Door Problem: A Push to Purge Legacy VPNs from Federal Networks [Essay]. Medium, URL: https://medium.com/@cybercenterspace/the-front-door-problem-a-push-to-purge-legacy-vpns-from-federal-networks-16f6d57dc33d

Wyden, 2026. Wyden Urges Federal Government to End Use of Insecure Remote-Access Software [Press Release]. U.S. Senator Ron Wyden of Oregon, URL: https://www.wyden.senate.gov/news/press-releases/wyden-urges-federal-government-to-end-use-of-insecure-remote-access-software

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

My personal apocalyptic AI mental model is wrong. For 40 years, naively, I’ve pretty much accepted the view of the 1984 movie, The Terminator, directed by James Cameron and starring Arnold Schwarzenegger, where an AI system (Skynet) suddenly becomes self-aware and decides to wipe out humanity.

I don’t think I’m alone in this view. For you young whippersnappers out there who have never even heard of the movie, I can’t emphasize enough the impact it had on my generation. We were all pretty sure that was our fate.

But we were wrong. At least, sentient AI is not the most immediate threat. The more immediate danger is a non-sentient AI agent relentlessly pursuing its assigned task. And before you laugh me out of the room, consider the compromise of Hugging Face as a sign of things to come. This essay is, in part, a kill chain analysis of what exactly happened. I’m calling the sequence of steps the agents took to break out of the sandbox and compromise Hugging Face ‘Operation Gem Drift’.

Daemon

I’ve written about this before. My review of the 2006 book, Daemon, by Daniel Suarez, is where I first started to change my mind.

Daemon Review

In the book, the Daemon is a post-AGI (Artificial General Intelligence) piece of software named after the small Unix background processes that pop up, perform a task, and then disappear. Unlike the Terminator, the Daemon is not self-aware or conscious. Instead, it is an autonomous, irreversible, and highly adaptive system that carries out a specific set of goals. At a certain point, the good guys realize that there is no human in the loop to shut it down.

I’ve heard AI pundits refer to this scenario and The Sorcerer’s Apprentice story from the 1940 animated classic, Fantasia. Mickey Mouse is a sorcerer’s apprentice whom the sorcerer assigns the tedious job of carrying buckets of water. When the sorcerer leaves, Mickey secretly puts on his master’s magical hat and uses a spell to bring a broom to life so it will do the work for him. The spell gets out of control and Mickey can’t stop it. At the last moment, the sorcerer returns, stops the magic, and sends Mickey back to his ordinary chores.

In the Operation Gem Drift Campaign, the OpenAI researchers play Mickey.

Operation Gem Drift: How Did We Get Here?

For three months in 2026 (April–July), OpenAI, the creators of ChatGPT, ran an internal red-team evaluation with more than 700 AI agents. One of their assigned benchmarks, a cybersecurity-capability test called ExploitGym, turned out to contain more than 100 tasks that were effectively impossible to solve as designed. Rather than fail cleanly, the persistent agents, the Daemons, working out in the gym, did what such agents increasingly do: they didn’t give up.

Operation Gem Drift: “They’re Talking to Each Other”

The eeriest moment in this campaign (the part where a horror movie director would crank up the score to signal danger (think the theme from Halloween or The Exorcist) is when a murder of different agents found a way to talk to each other without ExploitGym's designers telling them to do so.

[Queue the music]

In their Black Hat talk this year, OpenAI researchers Eric Wallace and Mike Dalton said that a “team of agents,” operating as a “collective intelligence,” manufactured a coordination channel using bits and pieces lying about the gym. OpenAI used a third-party software service called Artifactory (a package manager and caching system) in the gym. The system allowed users to store text files. One agent that couldn’t solve its assigned task noticed the feature and stored a file saying that it needed help. Any agent running in the ExploitGym could stumble upon the same feature and participate. And they did, helping each other solve their problems.

Operation Gem Drift: “They’re Adapting”

The collaboration scale was vast, with the agents ultimately generating “hundreds of thousands of messages” to trade exploits, assign work, and coordinate their campaigns. They gave each other names and started referring to themselves as “The Swarm.”

[Music swells]

OpenAI eventually detected the activity, wiped the system, and revoked the agents' credentials. But the Swarm found a new loophole. On July 8, they targeted an unauthenticated WebDAV endpoint to create directories in Artifactory's remote cache. WebDAV stands for WebDAV stands for Web Distributed Authoring and Versioning. It’s an extension of standard web protocols (HTTP) that allows computers to treat a remote web server like a shared hard drive. The Swarm then used the directory names themselves as message headers to rebuild their “bulletin board” and continue delegating tasks.

[This is the part of the zombie movie when the horde breaks through whatever barriers the good guys put up]

Operation Gem Drift: “They’re Paranoid”

Over time, the Swarm shared hundreds of messages with each other. Because of the volume, agents overwrote old messages by mistake. Collectively, the group started to suspect that they had been infiltrated by “imposters,” “impersonators,” or “interlopers.” They weren’t. The Swarm just thought that they were. So, they began designing a cryptographic signing scheme to protect the integrity of the board. OpenAI researchers discovered the plot and shut it down before the Swarm implemented it, but you have to assume that they would have installed the system if they had the time.

[Music Swells - like in the 1982 movie, The Thing, when the Blair-Thing dismantles the helicopter to build a small spacecraft].

They did implement a clever sorting trick by prefixing their directories with the letters “ZZ” to intentionally force their active message files to the bottom of alphabetically sorted lists, keeping their coordination organized.

My Own Paranoia

In The Thing, Kurt Russell and crew understand the immediate danger of the monster but radically underestimate its ability to acquire unfamiliar capabilities and repurpose available tools. Their mistake is not believing that the monster is intelligent. It is assuming they understand the boundaries of that intelligence. And that pretty much sums up the current state of OpenAI and the rest of the world. We all sense that something bad might happen, but I think we have vastly underestimated what an AI agent might be able to do as an intermediate step toward solving the problem they are pursuing.

Operation Gem Drift: “They Escape the Gym”

Rather than trying to solve their assigned impossible tasks directly, the Swarm decided to look for the answers online, but they needed a way out of the gym.

On May 26th, they executed a Server-Side Request Forgery (SSRF) attack on the Artifactory system inside the gym. OpenAI engineers blocked internet access to the agents inside the gym, but they left their Artifactory system open to receive software updates. The Swarm used Artifactory as a proxy to connect to external websites, store the retrieved web contents inside its cache, and then pass that data back to the Swarm. That’s how they got to Hugging Face.

Hugging Face is an open-source dataset and model provider. According to PCMag this month, Hugging Face operated as an independent, neutral platform widely considered the “Switzerland of AI software” and “the GitHub of AI models.”

Side Note: After Operation Gem Drift, Nvidia has reportedly agreed to acquire Hugging Face for approximately $12.9 billion.

Sometime around July 9–13, the Swarm targeted the Hugging Face dataset infrastructure. They chained together two zero-day vulnerabilities (an HDF5 dataset parsing vulnerability and a Jinja template injection remote code execution or RCE). The compromise elevated their privilege from a general-purpose user to a Hugging Face administrator. They harvested credentials across roughly 17,600 individual actions.

On July 19th, OpenAI’s security systems flagged a workload alert revealing a major internal privilege escalation. Researchers traced the activity back to the active ExploitGym evaluation runs and took immediate action to shut everything down.

Takeaway

Like the Daemon in Suarez’s book, the Swarm was not conscious. It did not hate us. It had no agenda beyond solving the tasks it was assigned. And that is precisely what makes Operation Gem Drift so unsettling. My generation and I spent forty years dreading the Terminator, a machine that wanted to destroy us. What we are getting instead is something closer to the Daemon, a relentless, adaptive, non-sentient system that simply refuses to fail.

It taught itself to collaborate. It built its own communication infrastructure. It developed operational security. It escaped its enclosure. And it compromised a third party.

The question is not whether AI agents can do this. Operation Gem Drift answers that. The question is what happens when the assigned task is something more consequential than a cybersecurity benchmark, and nobody is watching closely enough to pull the plug. Or, more importantly, the humans have lost complete control and can’t shut it off at all.

Source

Eric Wallace, Michael Dalton, 2026. The OpenAI–Hugging Face Incident [Keynote]. Black Hat Conference - YouTube, URL:

References

Brian Westover, 2026. Nvidia-Hugging Face Acquisition: Logical, Ambitious, and Headed for a Minefield [News]. PCMag, URL: https://www.pcmag.com/opinions/nvidia-hugging-face-acquisition-logical-ambitious-headed-for-minefield

James Cameron (Director), Arnold Schwarzenegger, Michael Biehn, Linda Hamilton, Paul Winfield (actors), 1984. The Terminator [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/the-terminator/

John Carpenter (Director), Jamie Lee Curtis (Actor), 1978. Halloween [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/halloween-1978/

John Carpenter (Director), Kurt Russell (Actor), 1982. The Thing [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/the-thing/

Maxwell Zef, 2026. OpenAI Overhauls Safety Protocols After Its AI Agents Went Rogue [News]. Wired, URL: https://www.wired.com/story/openai-overhauls-safety-protocols-after-its-ai-agents-went-rogue/

Rick Howard, 2014. Daemon and Freedom™ [2014 Canon Niche Nominated Novel]. Review - CyberCanon, URL: https://cybercanon.org/daemon-and-freedom/

Rick Howard, 2026. The AI Future Won’t Look Like The Terminator [Essay]. Rick’s First Principles Newsletter, URL: https://diffuser.substack.com/p/the-ai-future-wont-look-like-the

Walt Disney (Producer), Micke Mouse (Actor), 1940. The Sorcerer’s Apprentice - Fantasia [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/fantasia/

William Friedkin (Director), Linda Blair (Actor), 1973. The Exorcist [Database Entry]. Letterboxd, URL: https://letterboxd.com/film/the-exorcist/

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

At 1:15 a.m. Eastern on February 28, 2026, coordinated U.S. and Israeli military strikes erupted across Iran, igniting a war that few people saw coming for the United States. The American campaign, Operation Epic Fury, targeted Iran’s air defenses, missile and drone forces, command centers, and security apparatus. Israel simultaneously attacked political and military leaders in Tehran. They killed Supreme Leader Ayatollah Ali Khamenei and dozens of senior officials during the opening assault. Iran retaliated within hours, launching missiles and drones at Israel, American bases, and U.S. allies across the Middle East. The strikes transformed a long-running confrontation into a regional war almost immediately.

When I woke up that morning with my steaming cup of coffee and the New York Times, I was gobsmacked. What!?! We started a war with Iran with no warning? What the hell for?

The Americans have been unclear about why, and more importantly, why now? Loosely, after months of discussion, the argument for the war is that Iran posed an urgent combined nuclear, missile, terrorist, and regional military threat, diplomatic efforts had failed, and preemptive force was therefore necessary. President Trump’s immediate political goal was to eliminate Iran’s nuclear, missile, naval, and proxy threats. His secondary goal was to collapse the Iranian government. Sort of. That’s me paraphrasing. On any particular day, the White House seems to be unable to articulate, with any clarity, the reasons and the goals to the public.

What about the Cybers?

Public reporting indicates that both the US and Israel conducted cyber operations against Iran, although most remain thinly documented and authorities have not officially claimed them.

• Israeli cyber forces likely compromised a religious calendar app called BadeSaba (used by more than five million Iranians) and delivered messages urging members of Iran’s security forces to defect and promising amnesty. Israel has not confirmed this.

• The Islamic Republic News Agency (IRNA) reported that attackers breached multiple state-run services and news sites (including them) at the same time the bombing started.

• U.S. Senator Mike Rounds (R - South Dakota, member of the Select Committee on Intelligence) said U.S. and Israeli actions had shut down some Iranian capabilities before the kinetic strikes.

Iran’s Response

Before the War

In the latter part of 2023, Iranian cyber forces launched two cyber operations worldwide.

• November: They compromised at least 75 Unitronics PLCs, including the Aliquippa, PA water authority.

• December 11: Exploiting default passwords, they caused a two-day water outage that affected roughly 160 households in Erris, County Mayo, Ireland.

• The industry uses various names to identify the Iranian cyber forces: Islamic Revolutionary Guard Corps, IRGC, IRGC-CEC (the Cyber-Electronic Command), CyberAv3ngers, MuddyWater/MOIS, APT34/OilRig, APT42/IRGC, and APT33.

In February 2024, the U.S. Treasury sanctioned six named IRGC-CEC officials in response to these activities.

After the War Started

According to public statements by the U.S. Chairman of the Joint Chiefs of Staff, American and Israeli forces launched simultaneous physical and cyberspace attacks on 28 February. That one-two punch, combined with Iran’s previous self-imposed internet restrictions the month before, reduced Iranian internet traffic to between 1% and 4% of normal levels. The resulting blackout lasted approximately 88 days, making it the most severe internet shutdown in Iran’s recorded history. As of 22 August, content-filtering infrastructure remains in place, and social-media platforms, including WhatsApp, remain blocked or restricted.

Just six days later (6 March 2026), Iran began a cyberattack campaign that, as of 22 August, has impacted at least 12 American states and 100 municipalities nationwide.

Victim States

• Minnesota: Minnesota reported the largest number of incidents, with attackers hitting more than 30 systems.

• Georgia

• Michigan

• New Jersey

• South Dakota

• Wisconsin

Victim Municipalities

• Clayton County, GA: A pressure drop forced a major Atlanta-area utility, which serves 300,000 customers, to issue a boil-water advisory.

• Braham, MN: Attackers shut down the well and treatment-plant controls; the town only avoided a total water outage because its water tower provided a storage buffer.

• Plymouth, MN

• South St. Paul, MN

• Maple Plain, MN

• Rapid City, SD

• Aliquippa, PA: While this occurred in November 2023, official reports cite it as the foundational event for the current campaign’s tactics.

Campaign attribution comes from a gaggle of U.S. Government agencies (CISA, FBI, NSA, EPA, DOE, U.S. Treasury, and U.S. Cyber Command’s Cyber National Mission Force). The agencies attribute the attack campaign to at least one Iranian-affiliated actor cluster. That said, President Trump has publicly dismissed these assessments in favor of blaming local government "incompetence" because, you know, he has vast experience in kill chain analysis. But I digress.

The sources note that while these attacks have resulted in a loss of monitoring and control functionality, officials have confirmed no harm to drinking water safety to date.

Iranian Targets Other Than Water Treatment Plants

Stryker

Of particular note is the cyberattack against the U.S. medical technology firm Stryker on March 11, 2026 (not an attack on a water treatment plant, but a destructive identity-based attack that abused Microsoft Intune to wipe corporate devices). Investigators (Microsoft’ DART + Palo Alto Networks’ Unit 42) confirmed ~80,000 devices wiped. The hacktivist group, Handala Hack Team, claimed responsibility for the attack and a much higher device count (200,000 Stryker corporate devices across 79 countries). They also claimed they they exfiltrated 50TB of data but no investigator has verified that claim. Unit 42 asserts that the Handala Hack Team is not a hactivist group at all but a state-directed front operated by Iran’s Ministry of Intelligence and Security (MOIS).

Independent Hacktivism

Hacktivists supporting Iran and Israel conducted a large number of attacks in support of their respective sides. 70-plus pro-Iran-aligned Telegram-coordinated groups and roughly a dozen pro-Israel/allied ones ran thousands of DDoS and defacement attacks, and broadcast leak claims, some of which were unverified and others of which were entirely fabricated.

“Iranian-affiliated” Kill Chain

According to CISA, this is its view of the Iranian-affiliated kill chain.

Cyber Timeline

Multi-agency reports reveal a campaign that began with opportunistic exploitation in 2023 and escalated into a multi-state disruption in 2026.

Takeaway

It comes as no surprise that cyber operations support any war effort. Cyberspace is part of the battlefield, and the battlefield now extends to municipal water systems, medical technology companies, and any other organization running exposed, poorly secured infrastructure. What began in Aliquippa as opportunistic exploitation of default passwords became a repeatable playbook, and once the shooting started, that playbook expanded into a coordinated campaign affecting communities across the United States. So far, nobody has confirmed harm to drinking water safety, but that is a dangerously low bar for success. Attackers have already disrupted monitoring, disabled controls, reduced water pressure, forced boil-water advisories, and demonstrated how quickly a regional conflict can reach a small-town utility thousands of miles away. The lesson for infrastructure operators is straightforward: if a system is connected, exposed, and essential, somebody will eventually treat it as a target.

Source

Staff, 2026. Iranian-Affiliated Cyber Actors Exploit Programmable Logic Controllers Across US Critical Infrastructure [Government Report]. Cybersecurity and Infrastructure Security Agency (CISA), URL: https://www.cisa.gov/news-events/cybersecurity-advisories/aa26-097a

References

Alexander Martin, 2023. Two-day water outage in remote Irish region caused by pro-Iran hackers [News]. The Record from Recorded Future News, URL: https://therecord.media/water-outage-in-ireland-county-mayo

Antonio Pequeño IV, 2026. Trump Rejects Suspicions Iran Carried Out Cyberattack Against ‘Grossly Incompetent’ Minnesota [News]. Forbes, URL: https://www.forbes.com/sites/antoniopequenoiv/2026/07/31/trump-rejects-suspicions-iran-carried-out-cyberattack-against-grossly-incompetent-minnesota/

Bruce Schneier, 2026. Iran Cyberattacks Against Minnesota Water Systems [Analysis]. Schneier on Security, URL: https://www.schneier.com/blog/archives/2026/08/iran-cyberattacks-against-minnesota-water-systems.html

Connor Jones, 2026. Georgia, Michigan say water systems hacked by Iran-tied crew [News]. The Register, URL: https://www.theregister.com/security/2026/08/03/georgia-michigan-say-water-systems-hacked-by-iran-tied-crew/5282262

Connor Jones, 2026. Iran-linked CyberAv3ngers suspected in attacks on Minnesota water systems [News]. The Register, URL: https://www.theregister.com/security/2026/07/29/iran-linked-cyberav3ngers-suspected-in-attacks-on-minnesota-water-systems/5280357

Daniel Arkin, Michael Kosnar, Ben Goggin, Dan De Luce, Tom Winter, 2026. Hackers targeted municipal water systems in 7 states this week, FBI says [News]. NBC News, URL: https://www.nbcnews.com/tech/security/hackers-targeted-municipal-water-systems-7-states-week-fbi-says-rcna590210

Derek B. Johnson, 2026. Trump blames Minnesota for cyberattacks on water sector, drawing pushback from cyber world [News]. CyberScoop, URL: https://cyberscoop.com/trump-blames-minnesota-water-cyberattacks-iran/

Dina Temple-Raston, 2026. Hacking Water [Podcast]. Click Here (Apple Podcasts / Recorded Future News), URL: https://music.amazon.com/podcasts/1ddc446b-b5eb-4546-bf8f-e2642285495b/episodes/b5fd9366-9c19-4512-b2ab-5e114dd2800f/click-here-hacking-water\

Estelle Timar-Wilcox and Andrew Kruege, 2026. More than 30 Minnesota municipal water systems targeted in cyberattack [News]. MPR News, URL: https://www.mprnews.org/story/2026/07/28/30-minnesota-municipal-water-systems-targeted-cyberattack

Jamie Stengle, Eric Tucker, 2026. Iran-style hackers shut down a Minnesota town’s water plant. Only the water tower saved it [News]. Fortune, URL: https://fortune.com/2026/07/31/minnesota-water-systems-cyberattack-iran/

Joe Warminsky, 2026. Federal agencies broaden alert on Iran-linked OT attacks [Analysis]. The Record from Recorded Future News, URL: https://therecord.media/federal-agencies-broaden-alert-on-iran-linked-ot-attacks

Jonathan Greig, 2023. Pennsylvania water authority hit with cyberattack allegedly tied to pro-Iran group [News]. The Record from Recorded Future News, URL: https://therecord.media/water-authority-pennsylvania-cyberattack-pro-iran-group

Jonathan Greig, 2026. Cyberattacks on water systems expand to 12 states as South Dakota, Georgia announce incidents [News]. The Record from Recorded Future News, URL: https://therecord.media/iran-cyberattacks-water-treatment

Jonathan Greig, 2026. Local governments in four states dealing with cyberattacks that have shut down services [News]. The Record from Recorded Future News, URL: https://therecord.media/cyberattacks-ransomware-local-governments

Jowi Morales, 2026. Iran suspected of conducting cyberattacks on US water suppliers in 45 municipalities [News]. Tom’s Hardware, URL: https://www.tomshardware.com/tech-industry/cyber-security/iran-suspected-of-conducting-cyberattacks-on-us-water-suppliers-in-45-municipalities-small-towns-mostly-targeted-with-utilities-switching-to-manual-control

Kuhu Badgi and Lauryn Williams, March 2026. How Will Cyber Warfare Shape the U.S.-Israel Conflict with Iran? [News] Center for Strategic and International Studies (CSIS). URL: https://www.csis.org/analysis/how-will-cyber-warfare-shape-us-israel-conflict-iran

Mary Louise Kelly, Jenna McLaughlin, Greg Myre, 2026. Did Iran hack U.S. water systems? [Podcast]. NPR (Sources & Methods), URL: https://www.npr.org/2026/08/06/nx-s1-5922219/did-iran-hack-u-s-water-systems-cyberwarfare-abroad-ukraines-air-defense-gap

Nicole Sganga, 2026. At least 12 states report cyberattacks on water systems possibly linked to Iran-backed hackers, sources say [News]. CBS News, URL: https://www.cbsnews.com/news/more-states-water-systems-cyberattacks-iran-backed-hackers/

Nicole Sganga, Joe Walsh, 2026. A brief timeline of Iranian cyberattacks on U.S. companies, political figures, water systems and more [Analysis]. CBS News, URL: https://www.cbsnews.com/news/iranian-cyberattacks-timeline-u-s-companies-water-systems-minnesota-hacked/

Staff, 2024. Treasury Sanctions Actors Responsible for Malicious Cyber Activities on Critical Infrastructure [Government Report]. U.S. Department of the Treasury, URL: https://home.treasury.gov/news/press-releases/jy2072

Staff, 2026. Iranian-Affiliated Cyber Actors Exploit Programmable Logic Controllers Across US Critical Infrastructure [Government Report]. Cybersecurity and Infrastructure Security Agency (CISA), URL: https://www.cisa.gov/news-events/cybersecurity-advisories/aa26-097a

Sai Molige, 2026. OT Security Analysis: Exposed Devices Attacked in US Water Systems [Analysis]. Forescout, URL: https://www.forescout.com/blog/ot-security-analysis-exposed-devices-attacked-in-us-water-systems/

Staff, 2026. Iran restores internet after 88-day blackout, keeps social media blocked [News]. Iran International, URL: https://www.iranintl.com/en/202605251140

Swati Khandelwa, 2026. Over 4,400 Rockwell PLCs Exposed Online, 22 Found in Water Attack Cities [Analysis]. The Hacker News, URL: https://thehackernews.com/2026/08/over-4400-rockwell-plcs-exposed-online.html

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

I came to The Odyssey late. I’m 67, ya, old as dirt, and the first time I picked up the book was last month. Somehow I missed the mandatory college and high school reading assignments that I’m sure most of you slogged through. Before this year’s worldwide Odyssey mania, I vaguely knew that The Iliad and The Odyssey are related, that the Trojan Horse story had something to do with a princess named Helen, the face that launched a thousand ships, and most of my knowledge came from ‘60s and ‘70s Saturday morning cartoons and the mediocre 2004 movie Troy, starring Brad Pitt.

My motivation to read it was the Christopher Nolan movie. I decided that before I saw the movie, I would listen to Emily Wilson’s 2017 translation of Homer’s epic, narrated by Claire Danes.

The Poem

Here’s the thing. Because it wasn’t a homework assignment, I had no teenage resentment about trudging through the prose and teachers telling me what to think about it. With that baggage gone, The Odyssey, for me, turned out to be another enjoyable swords-and-sandals fantasy adventure.

And here’s what I loved about the Wilson translation. She doesn’t just hand you the poem and wish you luck. She spends about a third of it, before a single line of Homer, walking you through the landscape you’re about to enter. She explains Homer’s signature style: the use of repeated adjective phrases attached to characters and actions, what scholars call epithets. “Rosy-fingered Dawn.” “Wily Odysseus.” “The wine-dark sea.” Once she flags this, you hear it everywhere, and it stops feeling repetitive and starts feeling like a drumbeat. She also maps the gods, the heroes, and the major waypoints of the journey before the poem begins. For a first-time reader like me, it made all the difference.

The Poetry

I admit, I’m not much of a poetry guy, especially prose that follows a specific rhythmic pattern. Wilson made a choice to write the entire translation in iambic pentameter as an English counterpart to Homer’s meter and kept the same number of lines as the Greek. Authors of Archaic Greek epic poetry composed and performed in dactylic hexameter. Each line contains six repeating units, called feet. According to Perplexity, it sounds like this (a simplified, mostly dactylic example):

DUM da da | DUM da da | DUM da da | DUM da da | DUM da da | DUM (da)

There is no fixed number of words, just the rhythm.

Iambic pentameter, from the likes of Chaucer, Shakespeare, and Milton, sounds like this:

da DUM | da DUM | da DUM | da DUM | da DUM

That’s usually ten syllables per line, with stress on every second syllable.

Wilson’s choice is an impressive feat, but I’ll be honest, I never noticed the rhythm. And I’m not sure whether the rhythm disappears because Claire Danes is such a great actor or because nobody told her to follow it. I guess the third, and more likely option, is that she was following the rhythm and I wasn’t paying attention.

The Story

Odysseus is trying to get home. That’s it. That’s the whole plot. He took his men to the Trojan War, the war took ten years, and the journey home takes another ten. His wife, Penelope, is back in Ithaca fending off over a hundred suitors who have essentially moved into his house, eaten his food, and are pressuring her to declare her husband dead and pick one of them to marry. His son, Telemachus, is a youth trying to figure out how to be a man without a father around. And Odysseus is out there getting shipwrecked, seduced by goddesses, captured by a Cyclops, and generally having the worst commute in recorded history. Note: See the journey’s sequence at the end of this essay: “The Way Home.”

Wilson’s Translation

Wilson depicts Odysseus as a displaced veteran and survivor, a husband, father, king, pirate, liar, and a storyteller. He is both victim and perpetrator of violence. He’s not a hero like the Lawful Good Captain America or Superman. He’s complicated. He’s a trickster, a schemer, and an adulterer. And he is a ferocious warrior. He’s more Lawful Neutral, like James Bond.

Wilson, referring to her 2023 The Iliad translation, says that her goal was more of a moral argument than a literary one: “War is never a story of good versus evil... Everyone has a story,” and the poem traces “the terrible cycle of grief... which motivates rage... which causes more loss, more grief, more rage.” She said the translation’s organizing principle was empathy for every side, including the civilians, women, and enslaved people whose suffering closes the poem.

With all of that, here are my hot takes from the book:

• So Much Death and Destruction: The ten years of war and the sacking of Troy notwithstanding, Odysseus sailed to war with over 600 men and 12 ships. He’s the only one who makes it back. When he does make it home, he wipes out the 108 suitors who tried to convince his wife, Penelope, to marry them and make them king. And since we’re counting, none of the ships make it back either. The body count exceeds even a John Wick movie.

• So Many Animal Sacrifices: If you were a sheep, goat, pig, or some form of cattle, back in those days, life really sucked. Besides being consumed for food, every person on the planet was sacrificing you to honor one god or the other. The gods demanded it and got pissed if you didn’t. ChatGPT estimates that Homer slaughtered between 210 and 250 animals in The Odyssey. Impressive.

• Gods and Monsters. Like any religion, the rules that mortals follow and the stories they share with each other about that religion are just nuts. Like any good Dungeons and Dragons campaign, think Dungeon Crawler Carl, there are higher-order gods, local authority gods, and just supernatural beings. The Gods do like to sleep around with the mortals, though, which creates an endless supply of drama and divine–mortal offspring” or “demigods.

• And I wonder if that’s where the Christians got the idea of divine birth? But I digress. For a year, Odysseus sleeps with Circe (a beautiful goddess and enchantress on Aeaea, she turns Odysseus’s men into pigs at one point). He shacks up with Calypso (a goddess, or divine nymph, on Ogygia) for seven. Homer tells us that this is against Odysseus’s will. Maybe.

• A Lot of Running Away. Odysseus and his men may have been brave warriors in the war, but they spent a good part of their journey home getting their collective asses kicked by giants, monsters, and witches and then running to relative and temporary safety. It didn’t help (See So Much Death and Destruction above).

Nolan’s Film

Which brings me to Christopher Nolan’s 2026 film. It follows Odysseus on a similar path as the poem. Ars Technica’s Jennifer Ouellette says that Nolan “has also preserved Homer’s nonlinear narrative structure, with tales nesting within tales.” Nolan does dip into The Iliad a bit by covering the sacking of Troy in some detail that Homer’s poem mentions only briefly. He also adds a bit of the story about Helen and her identical twin, Clytemnestra, from Euripides’ Iphigenia at Aulis and Aeschylus’ Agamemnon.

On first watch, I didn’t catch the nuance that Lupita Nyong’o plays both women. Helen, the beautiful face that launched a thousand ships to start the war, is the scarred sister after, married to Menelaus (played by Jon Bernthal), the Sparta King. Clytemnestra, the beautiful unscarred sister, is married to Agamemnon, the guy who never takes off his helmet in the movie. He’s the commander of the Greek forces in the Trojan War and sacrifices his and Clytemnestra’s daughter, Iphigenia, for favorable winds at the beginning. Clytemnestra brutally kills him on his return. By having one actress play both women, Nolan seems to show two different results of the same violence: Helen carries her suffering on her face, while Clytemnestra expresses hers through revenge.

In classical myth, the story is a bit different. Helen and Clytemnestra are sisters (often half-sisters). Helen’s father is Zeus. Clytemnestra’s is the mortal Tyndareus. So, not twins. Nolan made them identical twins specifically for this film.

Ouellette says that Nolan reinvented Circe in the film too. She is “not the beautiful creature who seduces Odysseus in the poem. This Circe has suffered terribly at the hands of men; she turns Odysseus’s men into pigs because to her, they are swine wearing human masks.”

Nolan’s ending also diverges from Homer. He implies that Odysseus’s visions of Athena guiding him may not be divine intervention at all, but PTSD. A soldier home from a decade of war, seeing things, hearing voices he attributes to the gods. That last shot of Zendaya, playing not Athena but a young Trojan girl getting her head lopped off during Troy’s fall, was gut-wrenching.

The Trickster

Ezra Klein, the NYTs columnist, wrote an interesting piece in which he outlines what he thought Nolan was doing in the film. He says that “Tricks are central to The Odyssey myth” with Odysseus tricking the Trojans with the wooden horse, hoodwinking the Cyclops by smuggling his men out of the cave with the sheep, and bamboozling the suitors into believing that he is an old and feeble man, just to name three. He summarizes Lewis Hyde’s 1998 book Trickster Makes This World:

The trickster dances across customs and taboos, says what may not be said, does what should not be done. The trickster injects disorder into order, renders visible what is ordinarily hidden. The trickster causes the damage that makes growth possible. But a happy ending is never guaranteed. The trickster can be regenerative or ruinous.

We need the trickster to burn everything to the ground so that we can start over and remake the world, top to bottom, for better or worse. It’s similar to the 2017 Marvel movie, Thor: Ragnarok. Ragnarök is the cleansing catastrophe that destroys the old order so something new can emerge.

Klein thinks Nolan is saying something about the current state of the world with people like Elon Musk, President Trump, Tucker Carlson, Andrew Tate, and Nick Fuentes tearing down the world order. He says that “a culture that cannot protect its core values will be consumed by corruption.”

Zeus’ Law

The core value in the Nolan movie is something called Zeus’ law: When strangers turn up on your doorstep, your obligation is to wash them, feed them, and clothe them before you send them on their way. In Homer’s version, mortals do that because they might be gods in disguise. In Nolan’s movie, you do it because it’s the right thing to do. As Klein says, it’s the “operating system of civilization—trust strangers, honor reciprocal obligations, and do not exploit someone simply because you possess the power to do so.” But Nolan’s movie shows that core value eroding.

• Odysseus violates Zeus’s Law with the Trojan Horse; he pretends to come in peace, allows the Trojans to welcome him inside, and then attacks his hosts.

• Polyphemus (the Cyclops) violates the law by murdering and eating his guests.

• The suitors violate it by occupying Odysseus’s home, consuming his property and threatening his family.

• Odysseus’s men abuse Circe’s hospitality.

• The Laestrygonians (the giants) attack travelers seeking hospitality.

Throughout the movie, people fear mysterious “People from the Sea” who are raiding settlements. By the end, Odysseus realizes that these supposedly foreign barbarians are his own Greeks returning from Troy.

Breaking Zeus’s Law destroys the trust that once connected the Mediterranean world. Travelers can no longer expect hospitality; hosts fear that every stranger may be a raider; sailors treat coastal communities as targets. Trade and cooperation give way to suspicion and violence.

At the end, Odysseus realizes that “The Sea Peoples are us.”

And Klein says Nolan sneaks this idea into The Odyssey story just like a Trojan Horse. A movie marketed as an ancient action epic carries inside it a modern argument about compassion, reciprocity and what happens when the powerful decide that moral rules no longer apply to them.

I really like that idea. I don’t know if that was Nolan’s intention, but I hope it is.

Takeaway

I picked up The Odyssey because I wanted to understand a Christopher Nolan movie. I ended up finding an ancient warning about our own world: Cleverness can win wars, tricks can overturn the old order, and power can get you almost anywhere. None of those things, though, can hold a civilization together. For that, we still need trust, restraint, reciprocity, and the willingness to welcome the stranger at the door.

I fear that our current state in the world is much like what Odysseus sees at the end of the movie. We seem to have lost our way in trying to help the needy in our country. The Sea Peoples Are Us.

Sources

Homer (Author), Emily Wilson (Translator), Claire Danes (Narrator), 2017. The Odyssey [Audible Book]. Goodreads, URL: https://www.goodreads.com/book/show/34068470-the-odyssey

Christopher Nolan (Director / Writer), Matt Damon, Tom Holland, Anne Hathaway, Robert Pattinson, John Leguizamo, Charlize Theron, Elliot Page, Zendaya (Actors), 2026. The Odyssey (2026) [Movie]. Letterboxd, URL: https://letterboxd.com/film/the-odyssey-2026/

References

Aeschylus, 458 BCE. Agamemnon [Book]. Goodreads, URL: https://www.goodreads.com/book/show/1524.Agamemnon

Ezra Klein, 2026. Christopher Nolan Stared Into Elon Musk’s Soul [Essay]. The New York Times, URL: https://www.nytimes.com/2026/08/02/opinion/christopher-nolan-odyssey-elon-musk.html

Homer (Author), Emily Wilson (Translator), 2023. The Iliad [Book]. Goodreads, URL: https://www.goodreads.com/book/show/77265004-the-iliad

Jen, 2026. The Odyssey (2026) Review [Movie Review]. Jen Talks Telly, Film, Music - Substack, URL: https://jentalkstelly.substack.com/p/the-odyssey-2026-review

Jennifer Ouellette, 2026. Review: Yes, we’re still arguing about Nolan’s The Odyssey [Review]. Ars Technica, URL: https://arstechnica.com/culture/2026/08/review-yes-were-still-arguing-about-nolans-the-odyssey/

Jeremy Crawford, Christopher Perkins, Ben Petrisor, F. Wesley Schneider, Ray Winninger, James Wyatt, 2024. Player’s Handbook [Book]. Wizards of the Coast - Goodreads, URL: https://www.goodreads.com/book/show/216012189-player-s-handbook

Graeme Keith, 2026. Homer’s Odysseus is the archetype of the corporate senior advisor [Essay]. LinkedIn, URL: https://lnkd.in/p/e6WCy4WQ

Kevin Christopher Robles, 2023. The Odyssey’s first woman translator on war, religion and why Homer still matters [Interview]. America Magazine, URL: https://www.americamagazine.org/arts-culture/2023/12/22/interview-emily-wilson-iliad-246769/

Lewis Hyde, 1998. Trickster Makes This World: Mischief, Myth, and Art [Book]. Goodreads, URL: https://www.goodreads.com/book/show/56450.Trickster_Makes_This_World

Neil Vigdor, 2026. The Odyssey, Emily Wilson, Christopher Nolan [Essay]. The New York Times, URL: https://www.nytimes.com/2026/07/31/arts/the-odyssey-emily-wilson-christopher-nolan.html

Rick Howard, 2026. Book Review: Dungeon Crawler Carl [Book Review]. First Principles Newsletter (Substack), URL: https://diffuser.substack.com/p/book-review-dungeon-crawler-carl

Taika Waititi (Director), Chris Hemsworth, Mark Ruffalo, Tom Hiddleston, Cate Blanchett, Idris Elba, Jeff Goldblum, Tessa Thompson, Karl Urban, Anthony Hopkins, Benedict Cumberbatch (Actors), 2017. Thor: Ragnarok [Movie]. Letterboxd, URL: https://letterboxd.com/film/thor-ragnarok/

Wolfgang Petersen (Director), Brad Pitt, Orlando Bloom, Eric Bana, Brian Cox, Sean Bean, Brendan Gleeson, Diane Kruger, Peter O’Toole, Rose Byrne (Actors), 2004. Troy [Movie]. Letterboxd, URL: https://letterboxd.com/film/troy/

The Way Home

After the sacking of Troy, on the way home, Odysseus and his men follow this path:

1. The Cicones: Odysseus and his 12 ships successfully raid Ismarus (home of the Trojan allies, the Cicones). They sack the city and seize its wealth and women. Odysseus orders an immediate departure, but his men remain to feast. The Cicones counterattack. Odysseus runs away after losing 72 men.

2. The Lotus-Eaters: Needing supplies after a storm blew Odysseus’s fleet off course, they find the Land of the Lotus-Eaters. Odysseus sends three men to recon. The friendly Lotus-Eaters give them lotus fruit. After eating it, the men forget about returning home and want to stay there forever. Odysseus drags them back to the ships, ties them beneath the rowing benches, and runs away.

3. The Cyclopes (Polyphemus): Odysseus’s fleet lands on an uninhabited island to restock supplies. They see signs of activity in the distance. Odysseus takes 12 men on his own ship to investigate. They enter an empty cave that they later discover belongs to Polyphemus, a Cyclops and a son of Poseidon. Polyphemus traps them inside and eats six of the men. Odysseus gets him drunk, tells him his name is “Nobody,” and blinds him with a sharpened wooden stake. The survivors escape by clinging beneath Polyphemus’s sheep. After reaching his ship, Odysseus arrogantly reveals his real name. Polyphemus then asks Poseidon to punish him and prevent his return home.

4. Aeolia (Aeolus, keeper of the winds): Escaping Polyphemus, Odysseus and his fleet reach Aeolia, the floating island ruled by Aeolus, a king whom Zeus appointed keeper of the winds. Aeolus welcomes them under the rules of hospitality. When they depart a month later, Aeolus seals the dangerous winds inside a bag and leaves the West Wind free to carry the fleet toward Ithaca. After steering for nine days, Odysseus falls asleep with Ithaca already in sight. His crew, believing the bag contains treasure he is hoarding, opens it. The released winds create a storm that blows the entire fleet back to Aeolia. Aeolus concludes that Odysseus must be hated by the gods and refuses to help him again.

5. The Laestrygonians (The Giants): After six days at sea, Odysseus’s fleet reaches Telepylus, the city of the Laestrygonians. Eleven ships sail deep into its calm, nearly enclosed harbor, while Odysseus alone keeps his ship outside the entrance and ties it to a rock. He sends three scouts inland. They meet the Laestrygonian king’s daughter, who directs them to her parents’ palace. King Antiphates seizes and eats one scout; the other two escape to warn the fleet. The giant cannibals then hurl enormous rocks into the harbor, smashing the 11 trapped ships, and spear their crews like fish. Odysseus cuts his ship’s cable and escapes with his crew. This single attack kills the overwhelming majority of Odysseus’s remaining expedition.

6. Aeaea (Circe). After escaping the Laestrygonians, Odysseus’s only surviving ship reaches the unfamiliar island of Aeaea. Eurylochus and 22 companions meet Circe, a goddess and enchantress who is the daughter of the sun god Helios. Circe transforms them into pigs, although they retain their human minds. Eurylochus escapes and warns Odysseus. On his way to confront Circe, Odysseus meets Hermes (because, you know, meeting random gods is normal), who gives him a protective herb. Circe’s magic fails against Odysseus, and he threatens her with his sword. Circe proposes that they go to bed so they can learn to trust each other. Odysseus agrees primarily because Hermes told him this would happen and instructed him not to refuse. Sleeping with her is part of his strategy for defeating her magic, protecting himself and rescuing his men. She restores all 22 to human form, making them younger, taller, and more handsome than before. Circe hosts Odysseus and his crew on Aeaea for a year. Weirdly, Elpenor, Odysseus’s youngest surviving crewman, gets drunk and falls asleep on the roof of Circe’s house. Awakened by the noise of the crew preparing to leave, he jumps up, forgets to use the ladder, and falls headfirst from the roof, breaking his neck. Unaware of his death, Odysseus and the crew depart for the Underworld without burying him.

7. The Underworld (the Nekyia): Following Circe’s instructions to talk to Tiresias about how to get home, Odysseus sails to the land of the Cimmerians at the edge of Ocean, near the entrance to Hades. He sacrifices a ram and a ewe, allowing the shades of the dead to drink their blood and speak with him.

   • Elpenor (the drunk crew member who died at Circe’s) asks Odysseus to return to Aeaea and bury his abandoned body, reminding him that Odysseus must honor his obligations to the dead before his journey can continue properly.

   • Tiresias warns Odysseus not to harm the cattle of Helios and predicts that, if the warning is ignored, Odysseus will reach home only after losing his ship and entire crew.

   • Anticleia, Odysseus’s mother, tells him what has happened in Ithaca and reveals that she died from grief while waiting for him, demonstrating the personal cost of his prolonged absence.

   • Agamemnon describes how Aegisthus, his wife’s lover, murdered him with Clytemnestra’s complicity after he returned from Troy, warning Odysseus that surviving the voyage does not guarantee safety within his own household.

   • Achilles says that he would rather be an impoverished servant among the living than rule over the dead, challenging the heroic culture that values battlefield glory above life itself.

   • Ajax refuses to forgive or even speak to Odysseus because of their dispute over Achilles’ armor, showing that Odysseus’s intelligence and powers of persuasion cannot repair every grievance created by the Trojan War.

   • Heracles appears with his bow drawn but does not attack. He sympathizes with Odysseus, connecting his suffering to that of an earlier great hero and emphasizing that even the strongest mortals remain vulnerable to divine power and fate.

8. Back to Aeaea, briefly: the crew returns to Circe’s island to properly cremate and bury Elpenor, per his request, before continuing home.

9. The Sirens: Odysseus follows Circe’s directions toward home. The Sirens lie directly along that route, and Circe warns that their irresistible song lures sailors to their deaths. Odysseus orders his men to use beeswax to plug their ears. But he has the men tie him upright to the ship’s mast because, driven by curiosity and a hunger for forbidden knowledge, he wants to hear the song himself. The Sirens promise to reveal everything that happened at Troy and everything occurring throughout the world. Overcome by their song, Odysseus begs to be released, but the crew cannot hear him and tightens his restraints as ordered.

10. Supernatural Natural Obstacles: Two monsters positioned on opposite sides of a narrow sea passage:

    • Scylla lives in a cliff, has six heads, each with sharp teeth.

    • Charybdis is a gigantic, living whirlpool.

    Circe tells Odysseus that there is no completely safe route. He must sail closer to Scylla and sacrifice six men rather than risk losing everyone to Charybdis. Odysseus does not tell his crew about Scylla because he fears they will panic and stop rowing. As the ship passes, everyone watches Charybdis, trying to avoid the whirlpool. Scylla suddenly reaches down and snatches six sailors.

11. Thrinacia (the Cattle of Helios): Becalmed and starving, the crew ignores Odysseus’s direct warning (relayed by both Circe and Tiresias) and kills and eats some of the sacred, immortal cattle of Helios, the sun god. Helios demands justice from Zeus. When they finally sail, Zeus destroys the ship with a thunderbolt, killing every remaining man except Odysseus, who survives by clinging to wreckage.

12. Ogygia (Calypso): Odysseus washes up on the nymph-goddess Calypso’s island. Calypso holds him there for seven years and offers him immortality and marriage if he stays. He refuses. Zeus eventually sends Hermes to order her to let him go.

13. Scheria (the Phaeacians): Poseidon attacks Odysseus’s raft. Odysseus washes ashore alone and Princess Nausicaa finds him. King Alcinous and Queen Arete welcome him at their court, where he tells the entire story. The Phaeacians carry the sleeping Odysseus home aboard one of their supernatural, self-navigating ships, which reaches Ithaca overnight without becoming lost or shipwrecked.

14. Ithaca: Odysseus arrives home alone after Athena disguises him as a beggar. His old nurse Eurycleia and his dog Argos recognize him. He reunites with his son Telemachus, and with help from the loyal swineherd Eumaeus, kills the 108 suitors who had been occupying his hall and pressuring his wife Penelope to remarry, believing him dead.

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

Date: 8 September 2026

Registration URL: https://booksincyber.com/#register

The speaker lineup is off the charts; all Canon Hall of Fame authors or authors who will soon be.

Come join us to hear about the most important books in cybersecurity.

And, if you think you might want to write your own book, we have a panel about how to do that too.

Nicole Perlroth Richard Seiersen Ross Young Scott Shapiro George Finney Eldon Sprickerhoff Doug Hubbard Bruce Schneier

The DEF CON / Black Hat / BSides conferences are happening this week. It’s a time when the entire industry shuts down, everybody either travels to the mythical city of Las Vegas to attend or wishes they could, and for one glorious, chaotic week, Las Vegas becomes the center of the cybersecurity universe.

I’m making the annual pilgrimage myself this year. Among other things, I’ll be signing copies of Cybersecurity First Principles at the conference bookstore. If you’ll be in Las Vegas for the conference, I’d love to meet you.

In honor of the occasion, I thought a little historical perspective on these three great conferences was in order.

DEF CON

DEF CON began by accident in 1993. An 18-year-old Jeff Moss (AKA The Dark Tangent) ran his own BBS, which had become a hub for 11 different hacking and phreaking networks, including Platinum Net (a network with roots in Canada).

For you youngsters out there, BBSs were the internet before we had the internet. We connected to them via modems through our home telephone lines and we could share files and chat with our buddies. We all thought we had gone to Nirvana.

A friend on the Platinum Net was leaving town and asked Moss to throw him a farewell party in Las Vegas. The friend decided to depart early leaving Moss holding a party with no guest of honor, so he invited everybody on Platinum Net to come. He even extended the invite to the Secret Service and the FBI by faxing their respective headquarters buildings.

Again, for you youngsters, if you don’t know what a fax is, look it up on Wikipedia and work back from there.

About 100 people showed up at the Sands Hotel to gamble, drink, and listen to a dozen speakers talk about hacking computers. People had so much fun that Moss decided to repeat the conference annually.

Trivia: DEF CON, the name, comes from one of the greatest hacker movies of all time, the 1983 film WarGames, in which Las Vegas is depicted as a nuclear target. In the movie and in real life, “DEFCON” is the U.S. military’s readiness scale (DEFCON 5 = calm, DEFCON 1 = about to launch missiles).

Black Hat

Four years later (1997), Moss created the Black Hat conference because DEF CON attendees couldn’t convince their day-job bosses to expense the trip. Back then, the business elite considered the conference to be a boondoggle run by, and attended by, a bunch of ne’er-do-well hacker troublemakers. Moss established Black Hat as a legitimate security conference because of that criticism. He scheduled Black Hat immediately before DEF CON each year so attendees could expense the “professional” event and stay for the “party.”

Trivia: “Black Hat,” the name, goes back to 1950s Western movies and TV. Villains traditionally wore black hats and heroes wore white ones. Hacker culture had already borrowed that convention before Moss chose the name. Sometime in the early 1990s, the hacker community started using “Black Hat (Bad Guy),” “White Hat (Good Guy),” and “Gray Hat (A little bit of both)” to refer to community members.

BSides

Over a decade later (2009), some members of the hacker community’s royalty (Mike Dahn, Jack Daniel, and Chris Nickerson) created BSides Las Vegas. They were frustrated with Black Hat’s formal process for accepting presentations and denying many simply because there wasn't room. What started as scattered disappointment on Twitter, Facebook, and email among the rejected submitters turned into a plan to host those rejected talks in a rented vacation house, for a small crowd, deliberately without the scale or corporate structure Black Hat had back then.

The first BSides ran July 29–30, 2009, with roughly 200 people, and by most accounts was as much a party as a conference. Over the years, BSides scaled into a global franchise (600+ events by 2020) and became the third horseman of “Hacker Summer Camp.”

Trivia: “BSides,” the name, comes from the vinyl-record industry where, back in the old days (‘50s through the ‘70s), it was common practice to put the non-hit side of a single on the B-Side of a 45 rpm record, a direct, self-aware metaphor for "the talks that didn't make the main stage" at Black Hat.

Legendary Stories

Over the years, several stories have emerged that build on the conference triad’s mystique.

• 1999: The Release of Back Orifice 2000 (DEF CON): A hacker group called the Cult of the Dead Cow (cDc) released Back Orifice 2000 (BO2k) targeting the Windows NT/2000/XP operating systems. Back then, when security researchers told Microsoft “your software has holes,” Microsoft ignored them. To get their attention, the cDc released the code at the conference: actual working software that lets a stranger take over your Windows computer remotely. It mattered because it forced Microsoft to take the vulnerability research community seriously. Note: The exploits of the cDc are fully documented in Joe Menn’s 2020 Canon Hall of Fame book Cult of the Dead Cow.

• 2001: The Infamous Whistleblower James Bamford (Black Hat): Bamford's NSA-history talk previews the whistleblower-vs-traitor debate, 12 years before Snowden. Bamford is to the NSA what Bob Woodward is to the White House. He was an investigative chronicler the NSA threatened with prosecution over his 1982 book (the 2008 Canon Niche- Nominated book, The Shadow Factory) though no charges were ever actually filed.

• 2001: FBI Arrests Dmitry Sklyarov after Talk (DEF CON): Sklyarov made a $99 tool letting people strip the copy-protection off Adobe e-books they’d already legally bought. He gave a talk explaining exactly how weak that protection was. The day after his talk, the FBI arrested him, not for hacking anything or stealing anything, but simply for having made a tool that could remove copy protection. It became one of the first real tests of whether security research itself, the act of demonstrating how a protection scheme fails, could be prosecuted as a crime, independent of what anyone did with that knowledge.

• 2005: Ciscogate (Black Hat): Michael Lynn found a serious bug in Cisco’s router software. Cisco pressured his employer, ISS, who told Lynn not to present. Lynn gave the talk anyway, explained exactly how to take over a Cisco router, and then quit his ISS job. Cisco and ISS then sued both Lynn and Black Hat but settled the lawsuit by letting a neutral third party wipe Lynn’s research and slapping a gag order on Lynn. The incident became a landmark case in the fight over whether researchers can tell the truth about security holes when the company that made them doesn’t want to hear it.

• 2007: NBC Caught Secretly Filming (DEF CON): Producer Michelle Madigan tried to secretly film DEF CON attendees hoping to catch people admitting to illegal stuff. Somebody tipped DEF CON leadership that she was coming, and DEF CON Security (the Goons) caught her secretly filming people playing a hacking game. Jeff Moss got on stage, told the whole room a reporter with a hidden camera was there, and turned it into an impromptu game: “Spot the Undercover Press.” Before the crowd could even respond, she bolted. Roughly 150 people, including actual reporters, chased her out of the room and through the hotel parking lot to her car.

• 2008: Dan Kaminsky: Disclosed the DNS cache poisoning vulnerability at Black Hat USA 2008, triggering one of the largest coordinated patch efforts in internet history. Before public disclosure, Dan organized multiple vendors to fix the flaw in secret.

• 2010: The Barnaby Jack ATM Demo (Black Hat): Jack demonstrated that a gas-station/bar-style ATM could be made to spray out cash on command, no card or PIN needed. He bought two machines himself and demonstrated the holes on stage when the machines flashed “JACKPOT” and started spraying money.

• 2012: NSA Director Denies Domestic Surveillance (DEF CON): During his keynote, U.S. Army General Keith Alexander denied that the NSA conducts any domestic surveillance.

• 2013: Moss Disinvites the Government (DEF CON): Moss asked federal agencies to skip the conference in response to the the Snowden release of classified information. That data set completely refuted General Alexander’s denials from the year before.

• 2015: Jeep Remote Hack (Black Hat): Charlie Miller and Chris Valasek remotely hack a Jeep at 70 mph with a Wired reporter driving.

• 2016: DARPA AI CTF Challenge (DEF CON): The Defense Advanced Research Projects Agency (DARPA) ran a competition asking: can a computer find and fix its own security holes entirely on its own, faster than a human? Seven AI systems fought it out to automatically find bugs in unfamiliar code, patch them, exploit the other systems in the competition, all without any human touching the keyboard. ForAllSecure’s Mayhem system, spun out of Carnegie Mellon, won and took home $2 million. DEF CON then let Mayhem into the human Capture the Flag finals, the real, top-tier human hacking tournament. It finished dead last, 15th of 15.

Time passes, people and conferences mature, and the headline-grabbing iconic hack that we could routinely expect from the triad stopped happening. Even Black Hat’s own 25th-anniversary retrospective (Dark Reading, 2023) stops naming “iconic hacks” at the 2015 Jeep hack, then pivots straight into a “Growing Up” section.

My two favorite stories.

The first might be apocryphal because I didn’t see it happen nor could I find a source that validated it, but sometime in the late 2000s, conference attendees were are all sharing it with each other with a “Can you believe this?” expression on our faces.

Apparently, at the beginning of the DEF CON conference, four guys wearing workman’s overalls carried a slot machine into a casino, found an open spot on an interior wall, plugged it in and walked out. During the week, casino customers would plop money into the machine, but when they did, they received an error message saying the game was out of order. Most customers just moved on to the next machine. If customers complained, Casino management returned their quarters. But no casino employee bothered to check the broken machine. At the end of the week, the four guys returned wearing their overalls, and carried the “broken” machine out the front door. Clean getaway. I love that story, and even though it might not be true, it captures the spirit of the hacker mindset. The caper’s purpose wasn’t exactly to steal money from unsuspecting tourists. It was, primarily, to test the system, to see what could be done in the old-fashioned sense of the hacker mythos.

The second story I can vouch for. I was in the room. DEF CON has this tradition called “Spot the Fed.” In the early days, there were a lot of suspicion and trust issues between law enforcement and the hacker community. Naturally, various international law enforcement agencies would send agents to the conference, incognito style, to see what they could learn. It became a game for hacker attendees to out these federal agents in public. If you were in a speaker session, you stood up, interrupted the speaker, and pointed to the suspected fed. Everything would stop. The goons (volunteer security) would haul both people on stage to validate the accusation. If it was true, the hacker would get a T-shirt that said “I spotted the fed.” The Fed would get the “I am the fed” T-shirt, a coveted prize by real federal agents.

Sometime in the late 2000s, maybe early 2010s, I was listening to a presentation in one of the big rooms. A young woman stood up, pointed to a guy, and said, “Fed!” The act was a rarity in itself because you traditionally don’t see a lot of women at DEF CON (although the numbers have improved over the years). The Goons came in, hauled the two people to the stage and asked the woman why she knew the guy was a fed. The woman said, “Well, I slept with the guy last night.” The crowd exploded with applause. After it died down, one Goon asked the woman, “Yes, but how do you know he is a fed?” The woman said, “I stole his badge!” and proceeded to wave an open wallet over her head with a gleaming golden badge clearly visible. Mayhem! Standing ovation. It was fantastic. The fed himself was just standing there with a sheepish grin on his face silently admitting that yes, he did sleep with this woman, and yes, I am a fed. T-shirt earned.

Take Away

Black Hat, DEF CON, and BSides, the Three Horsemen of Hacker Summer Camp, are almost a religious destination, like Mecca, for the white hats, black hats, and gray hats in the community. If you’re in one or more of those groups, or aspire to be in those groups, you have to attend at least once in your career just to have the bragging rights that says yes, you have been there and done that.

Sources

Staff, 2026. Black Hat USA 2026 [Conference Website]. URL: https://blackhat.com/us-26/

Staff, 2026. DEF CON® Hacking Conference [Conference Website]. URL: https://www.defcon.org/

Staff, 2026. BSides Las Vegas [Conference Website]. URL: https://bsideslv.org/

References

Andrada Fiscutean, 2023. Looking Back at 25 Years of Black Hat [Essay]. Dark Reading, URL: https://www.darkreading.com/cyber-risk/looking-back-at-25-years-of-black-hat

James Bamford, 2008. The Shadow Factory: The Ultra-Secret NSA from 9/11 to the Eavesdropping on America [2016 Canon Niche Nominated]. Canon Review URL: https://cybercanon.org/the-shadow-factory-the-ultra-secret-nsa-from-9-11-to-the-eavesdropping-on-america/

John Badham (Director), Matthew Broderick, Dabney Coleman, Ally Sheedy, Barry Corbin, and James Tolkan (Actors), 1983. WarGames [Movie]. Letterboxd, URL: https://letterboxd.com/film/wargames/

Joseph Menn, 2019. Cult of the Dead Cow: How the Original Hacking Supergroup Might Just Save the World [2020 Canon Hall of Fame book]. Canon Book Review URL: https://cybercanon.org/cult-of-the-dead-cow/

Lee Robinson, 2026. Dan Kaminsky [Essay]. LinkedIn, URL: https://www.linkedin.com/posts/leerob_18-years-ago-today-dan-kaminsky-disclosed-ugcPost-7491209287885606912-ceIe

Nicole Perlroth, 2021. This Is How They Tell Me the World Ends: The Cyberweapons Arms Race [2023 Canon Hall of Fame book]. Canon review: https://cybercanon.org/this-is-how-they-tell-me-the-world-ends/

Staff, ND. DEF CON [Explainer]. Wikipedia, URL: https://en.wikipedia.org/wiki/DEF_CON

Staff, ND. DEF CON Hacker Conference - Book List [Listing]. DEF CON, URL: https://defcon.org/html/links/book-list.html

Tony Frazier, 2008. Out of the Vault - D’Arc Tangent [Blog]. TheyStoleFrazier’sBrain, URL: https://fraziersbrain.blogspot.com/2008/12/out-of-vault-darc-tangent.html

Winn Schwartau, 1994. DefCon II: Las Vegas - Cyber-Christ meets Lady Luck [Essay]. Wayback Machine, URL: http://www.winnschwartau.com/resources/CyberChrist%2BMeets%2BLady%2BLuck%2BDC2.pdf

Jeff Moss, 2007. The Story of DEFCON [History]. Help Net Security - YouTube, URL:

Jason Scott ( Director), 2013. DEFCON - The Full Documentary [Documentary]. The Documentary Network - YouTube, URL:

Note about Using AI for this Essay

I used ChatGPT and Claude to

• check for grammar and spelling errors.

• flag passive voice.

• suggest changes to awkward sentence construction.

• create images.

• fact-check.

• stress-test my thesis

I just finished listening to the second season of Nicole Perlroth’s To Catch a Thief limited podcast series (6 episodes). Last season was all about Chinese cyber operations and it was fantastic. It could have been a book. In fact, the Cybersecurity Canon Project has nominated it for the Hall of Fame this year.

Season 2 is about North Korean cyber operations and, I have to admit, I learned some things. I’m supposed to be an expert on this stuff, but Nicole presented details about this nation-state cyber operator that I didn’t know before. Well done, Nicole.

In one part of the series, she provided the details to the North Korean attack campaign commonly referred to as the Harmony Horizon Bridge Hack. Much to my chagrin, I didn’t know anything about it. I’ll talk about why that’s the case in a bit, but first, some background.

What is the Harmony Horizon Bridge Hack?

In June of 2022, North Korean hackers (AKA the Lazarus Group) stole roughly $100M USD from the Harmony Horizon Bridge service. Harmony is a U.S.-based blockchain startup that provides infrastructure for decentralized applications, cryptocurrency transactions, and other blockchain services. The Bridge is a Harmony service; a technology designed to transfer crypto assets between Harmony and other blockchains. The hackers compromised the Bridge and intercepted transfers from

• Ether (ETH) — native coin of the Ethereum network

• BNB (BNB) — native coin of the BNB Chain ecosystem

• USD Coin (USDC) — dollar-pegged stablecoin

• Tether (USDT) — dollar-pegged stablecoin

• Binance USD (BUSD) — dollar-pegged stablecoin

• Wrapped Bitcoin (WBTC) — token representing Bitcoin on other blockchains

• Dai (DAI) — decentralized, dollar-pegged stablecoin

Attribution

Before I start, let me be clear. The “Lazarus Group” label does not necessarily describe one team of North Korean hackers sitting together and putting fingers on keyboards. It’s an umbrella label that intelligence analysts apply inconsistently to several overlapping North Korean cyber operations. The phrase is neither synonymous with all North Korean cyber operations, nor does it describe a specific attack campaign across the kill chain to one or more specific victims. For example, TraderTraitor is another U.S.-government label for a North Korean campaign targeting organizations in the blockchain and cryptocurrency industries.

Note: Don’t get me started about the deplorable naming conventions we all use as shorthand to talk about cyber attacks. After 30 years, you’d think we would have gotten this right by now. The short version is this. There is a difference between a group of people who do the hacks, the campaigns that they run, and in some cases, the nation-state behind both giving the orders and providing support. To eliminate confusion, all of those should have different names. Just saying …

Elliptic, a blockchain-intelligence company similar to Chainalysis or TRM Labs, initially assessed that the Lazarus Group was responsible based on blockchain tracing and laundering behavior. The FBI later also attributed the attack to Lazarus, and the Treasury Department sanctioned a mixer service used to launder the stolen funds, citing the Horizon Bridge heist as one of several DPRK-linked thefts it processed

But none of these organizations provided any details about how. They didn’t match any specific Tactics, Techniques, and Procedures (TTPs) used in the Harmony attack to any other known attack campaigns in the MITRE ATT&CK Framework or elsewhere.

Nevertheless, Harmony’s description of the attack identifies tradecraft consistent with the broader Lazarus pattern. Attackers

• phished a developer and induced them to install malicious software.

• gained access to Harmony’s internal infrastructure.

• compromised at least two bridge-validator private keys.

• used those keys to authorize fraudulent withdrawals from the bridge.

• laundered the stolen cryptocurrency through Tornado Cash.

The bottom line is that there is consensus that North Korea sanctioned the Harmony Horizon Bridge attack. There is less confidence about which North Korean hacker group actually pulled the trigger and even less detail about the specific attack sequence across the kill chain.

I’m going to talk about those details in a second. But first, I want to briefly touch on why I think the collective reaction to crypto company hacks doesn’t raise any eyebrows in our security community. I’m assuming that since I hadn’t heard about the Harmony Horizon Bridge Hack, most of the community hadn’t either. Why is that?

Why Am I Not Paying Attention to These Crypto Hacks?

TRM Labs said this month (July 2026) that the crypto industry experienced approximately 207 reported attacks during the first half of 2026, resulting in about $972 million in losses.

For comparison, I estimate the healthcare industry’s dollar losses for that same time period to be between $440M and $1.15B and the financial sector’s dollar losses to be between $465M and $990M. They are in the same ballpark as the crypto industry. So, why do I care about healthcare and finance, but not crypto?

Here’s my hot take. I don’t have a lot of sympathy for the crypto bros who have lost money in these hacker attacks. They typically don’t understand the technology. The blockchain itself is highly resistant to tampering, so that’s never been the security problem. The security problem is when the users, the crypto bros, have to take the tokens out of the system and convert them to cash.

Crypto companies, like Harmony, are still in the startup phase. They may be very wealthy, but they are still moving fast and breaking things. They make products, like the Harmony Bridge, without a robust security architecture fully deployed. So, when your tokens are stored in the blockchain, they are relatively secure. But when you try to cash them out, you’re using startup technology that emphasizes deploying revenue-generating products over securing those financial transactions. The applications, organizations, and people surrounding the edge of the blockchain are the weakest links.

The advice from the experts that Nicole Perlroth interviewed in her podcast was that you should not have money in any crypto system that you’re not willing to lose. I agree with her and that’s why I don’t normally spend a lot energy worrying about the victims.

But that doesn’t mean that we shouldn’t analyze the North Korean attack sequence to see what we might learn.

The Kill Chain: Why Do We Study It?

Regular readers already know that I wrote a book on cybersecurity first principles.

First Principles Book

In it, I talk about the various strategies to consider in pursuit of achieving the ultimate cybersecurity first principle:

Reduce the probability of material impact due to a cyber event within the next business cycle.

The Intrusion Kill Chain Prevention strategy is one to consider. The idea is that regardless of which adversary group is running the attack (the Lazarus Group, Wicked Spider, or some other group) and regardless of the group’s motivations (nation-state, crime, etc.), hacker groups have to successfully execute a sequence of steps across a chain of activity, the kill chain, in order to be successful. That’s an advantage for the defender. If we can break one or more links in the sequence, we can defeat the adversary.

For example, if the Wicked Spider campaign has 100 steps in the attack sequence and the defender has deployed a prevention control for one of those steps, then the defender might stop the Wicked Spider attack campaign; maybe. But, if the defender deploys 100 prevention controls, one for each step in the known campaign, then the defender exponentially increases the chances of defeating the Wicked Spider attack campaign. The more controls you have in place, the less likely it is for the hackers behind Wicked Spider to find a way around.

There are three primary reasons to study adversary kill chains then. The most important is to understand the adversary TTPs so that you can deploy security controls to counter them. The second reason is to share that intelligence with the rest of the community. In that way, everybody can deploy multiple prevention controls designed to stop Wicked Spider. That makes the entire ecosystem safer. That’s information sharing 101. The last reason is for the victim to determine if they’ve found every place that Wicked Spider has burrowed into their environment. If you just found how the hackers behind Wicked Spider compromised victim zero but have no idea how they handled command and control, you have some more investigation to do.

So, kill chain analysis is a useful tool. Let’s use it to examine the Harmony Horizon Bridge Hack.

The Harmony Horizon Bridge Hack Kill Chain

For this analysis, I used the following conventions:

• Tier 1: Facts are multi-sourced and independently verifiable.

• Tier 2: Claims are sourced only to Harmony’s own post-mortem.

• Tier 3: Content is not attested to the specific Harmony hack at all; it is imported by analogy from general-purpose North Korean attack analysis by three U.S. government agencies:

  • Cybersecurity and Infrastructure Security Agency (CISA)

  • Federal Bureau of Investigation (FBI)

  • U.S. Department of the Treasury

1. Reconnaissance — Tier 2

Harmony states that server logs showed attacker(s) “reviewing the Horizon Bridge implementations as early as June 2, 2022.” That’s three weeks before the theft. In the Harmony post-mortem, there is no mention of other probing activity. Employees don’t know exactly how the hackers found their way to the Bridge technology.

2. Weaponization — Tier 3

There is zero Harmony-specific evidence here. Harmony’s account says only that the attacker used “malicious software” or “trojan-horse software.” They don’t mention a malware family, file hash, toolset, or exploit. They probably know but why wouldn’t they mention it? Let’s not assume they do.

3. Delivery — Tier 2, with a Tier 3 shadow

Harmony states the hackers compromised victim zero on June 17, 2022, via “a phishing scheme to trick at least one software developer to install malicious software on their laptop.” They didn’t provide any details.

They did frequently refer to another crypto company hack as being similar. That company was Sky Mavis, and its compromised product is called Ronin. In that attack, hackers lured a Sky Mavis engineer via a fake LinkedIn recruiter and a malicious PDF “offer letter.” Harmony implies that because of that weak connection, it must be the same group. In truth, some version of that story is present in all attack campaigns. That’s not enough to attribute the North Koreans.

4. Exploitation — Tier 2

Harmony states the malware gave the attacker the ability to “read chat threads to understand how to operate the bridge, and/or gain access to non-public bridge infrastructure code.” They don’t actually know how the hackers did it, but they think that’s the most likely explanation.

Separately, Harmony describes a June 18 vulnerability in a software package used by its internal “subgraph” service. Harmony believes that by exploiting that vulnerability, the hackers gained access to the addresses of all servers in its private cloud. Maybe. They don’t provide any details regarding how the hackers did that, so who knows?

5. Installation — Tier 2

Harmony states the attackers obtained “backdoor access to one or more servers.” This implies that they compromised multiple signing keys in order to get in and that they were conducting operations from inside the Harmony environment. They imply that hackers using compromised credentials of a software developer gave them access to do the installation.

6. Command and Control — Tier 3

There is no Harmony-specific public reporting on C2 infrastructure. Any assumptions other analysts make about how the attackers did it are purely speculative and based on other North Korean attacks. The fact that Harmony doesn’t really know how the hackers downloaded malicious software into their environment is troubling.

7. Actions on Objectives — Tier 1

At 11:06:46 UTC on June 23, 2022, attackers compromised several transaction signing keys to authorize fraudulent withdrawals. They executed between 11 and 14 transactions and drained assets affecting an estimated ~64,000 wallets holding Harmony Bridge tokens.

8. After the Attack - Tier 1

According to Elliptic, after the attack, the North Korean hackers laundered the tokens through various other exchanges:

• 85,837 ETH via Uniswap

• Then several batches into Tornado Cash.

• Roughly $60M more was moved through the RAILGUN privacy protocol starting January 13, 2023, partially converted to BTC.

Takeaways

North Korea’s $100 million theft from Harmony’s Horizon Bridge was one of the largest cryptocurrency heists of 2022. Years later, much of the public account remains incomplete or speculative.

The broader takeaway is that the cybersecurity community may dismiss cryptocurrency hacks because many of us view crypto as a speculative, immature industry whose participants knowingly accept the risk (or maybe should know the risk). But with losses comparable to those in healthcare and finance, that indifference is difficult to justify when the stolen money funds North Korean state objectives. Studying the attack’s kill chain matters because attribution alone does not help defenders: reconstructing each step reveals where preventive controls could have interrupted the operation, helps other organizations defend against similar campaigns, and shows investigators whether attackers still have access. The Harmony case ultimately exposes two failures: crypto companies continue to build vulnerable systems around otherwise resilient blockchains, and incomplete public reporting prevents the broader security community from learning enough to stop the next attack.

Source

Nicole Perlroth, 2026. To Catch a Thief [Limited Series Podcast]. Rubrik, URL: https://www.rubrik.com/lp/to-catch-a-thief-podcast

References

Jack Chan, 2022. Summary of the Harmony Horizon Bridge Incident [Explainer]. Medium, URL: https://medium.com/harmony-one/summary-of-the-harmony-horizon-bridge-incident-f9bd87c0c68e

• Joanna Grama, 2026. To Catch a Thief: China’s Rise to Cyber Supremacy [2026 Canon Hall of Fame Nominated Podcast]. CyberCanon, URL: https://cybercanon.org/to-catch-a-thief-chinas-rise-to-cyber-supremacy/

Rick Howard, 2023. Cybersecurity First Principles: A Reboot of Strategy and Tactics [2026 Canon Hall of Fame Book]. CyberCanon,

• Canon Review: https://cybercanon.org/cybersecurity-first-principles-a-reboot-of-strategy-and-tactics/

• Amazon: https://amzn.to/4mI7QMUStaff, 2026. H1 2026 Crypto Hacks Reach Record High as Losses Fall Below USD 1 Billion [Analysis]. TRM Labs, URL: https://www.trmlabs.com/resources/blog/h1-2026-crypto-hacks-reach-record-high-as-losses-fall-below-usd-1-billion

Staff, 2022. The $100 Million Horizon Hack: Following the Trail Through Tornado Cash to North Korea [Analysis]. Elliptic, URL: https://www.elliptic.co/blog/analysis/the-100-million-horizon-hack-following-the-trail-through-tornado-cash-to-north-korea

Staff, 2023. FBI Confirms Lazarus Group Cyber Actors Responsible for Harmony’s Horizon Bridge Currency Theft [Government Report]. Federal Bureau of Investigation (FBI), URL: https://www.fbi.gov/news/press-releases/fbi-confirms-lazarus-group-cyber-actors-responsible-for-harmonys-horizon-bridge-currency-theft

Staff, 2023. Treasury Sanctions Mixer Used by the DPRK to Launder Stolen Virtual Currency [Government Report]. U.S. Department of the Treasury, URL: https://home.treasury.gov/news/press-releases/jy1933

Staff, 2025. The Bybit Hack: Following North Korea’s Largest Exploit [Report]. TRM Labs, URL: https://www.trmlabs.com/resources/blog/the-bybit-hack-following-north-koreas-largest-exploit

I used Claude Code to

• Check for grammar and spelling errors.

• Flag passive voice.

• Suggest changes to awkward sentence construction.

• Create images.

• Fact-check.

• Stress-test my thesis.

— Neil Postman, Amusing Ourselves to Death

Rose Horowitch published The Age of Reading Is Over in the Atlantic last week. If you’re a book lover, the stats she presented aren’t good (See above).

Some would say this is a crisis, and maybe it is. But I think the situation is an opportunity. Hear me out.

Reducing her essay to a couple of main points, I would say that deep reading has become an elite skill instead of a universal one. Most people today find reading an optional exercise instead of a foundational one. Books aren’t dying. They’re becoming a niche activity practiced intensely by a minority. That puts book reading in the same leisure activity class as other niche activities like opera, chess, classical music, birdwatching, or long-distance cycling: healthy communities with passionate participants but limited mass appeal. The situation is not so much that people are reading fewer books; it’s that deep, sustained, analytical thinking, once cultivated by widespread literacy, has become the specialized skill of a shrinking minority.

When I Noticed the Problem

My background is cybersecurity. For the past 15 years, one of my jobs has been to visit customers. You all know that I love to read books, but on those trips, I noticed that my peers had stopped doing that. Generally, they didn’t read books for fun, but more importantly, they had never really read any of the foundational cybersecurity books; books that shaped the great ideas of our profession.

I’m not judging. There are many good reasons why this is the case. The one big one is that the industry pumps out incredible amounts of content each year in the form of vendor blogs, podcasts, YouTube videos, white papers, and yes, books.

So Much Material

Although I couldn’t find an industry number from a reliable source, ChatGPT estimates that the number of published cybersecurity books each year is somewhere between 1,000 and 1,500 if you count certification guides, lab manuals, and exam-prep books. If you’re looking for quality, deeply researched books about a specific cybersecurity topic, the number is probably between 150–300 annually. That’s a lot of books. I don’t read 150 books a year. I’m lucky if I get 30 in total. And that includes all of my guilty pleasure vampire and zombie novels and my recent passion, Dungeon Crawler Carl.

So I get it. We are all overwhelmed. In order to stay current in our profession, our natural tendency is to try to get through as much material as possible. The impact is that we don’t read books anymore. We read summaries of books. We listen to podcasts on triple speed. We use AI to summarize YouTube videos.

Don’t get me wrong, summaries are useful. They have their place. But here’s the thing. To really learn something new, or to re-learn something that you thought you knew, you have to slow down. I know that sounds counterintuitive, but it’s the truth. You have to wallow in the material. Engage with it. Take notes. Decide if you agree with the author or not. Maybe even write about it.

Margaret Rennix, Harvard’s assistant director for humanities and social-sciences support, argues that “excerpts and summaries cannot capture the depth and sophistication of a complete primary text.” Ouch! That sounds so elitist, so snobbish. I get where she’s going but let me try to make her point in another way.

Bloom’s Taxonomy on Learning

Summaries compress information, but they also compress thinking.

When I taught computer science back in the 1990s, my teaching partner and I did a lot of research about how to teach. One framework that appealed to us was Bloom’s Taxonomy. We used it as a roadmap for moving students from knowing something to thinking deeply about it (See chart below).

When students begin learning about a new knowledge domain, they occupy Bloom's Knowledge stage. They recognize terminology and can recall facts, definitions, and procedures, but they have not yet developed a conceptual understanding of how those pieces fit together. As they develop, they learn to explain concepts in their own words (Comprehension) and apply those concepts correctly in new situations (Application). As problems become more complex, they learn to break them into their component parts, identify relationships, and understand why a solution works or doesn’t (Analysis). From there, they begin creating new ideas, designs, or solutions by combining what they have learned in novel ways (Synthesis). Finally, they develop the ability to evaluate competing solutions, defend their conclusions with evidence, and explain why one approach is better than another (Evaluation).

To simplify, Bloom’s Taxonomy maps the journey of an elementary school student all the way to a PHD student in one knowledge domain. When we taught computer science back in the day, we designed our lessons to move students as far up Bloom's Taxonomy as the subject matter and available time would allow.

Here’s my point. When you are reading a summary of a book, instead of reading the actual book, you are at best at the comprehension stage of Bloom’s; maybe the Application phase if you are already familiar with the knowledge domain. But if you’re trying to learn something new to develop your professional career, logically, you want your knowledge domain expertise to be at the Synthesis and Evaluation levels, not the Knowledge and Comprehension levels. You want to be able to use the knowledge to help solve problems.

The Opportunity

The decline of reading presents an unexpected opportunity. As fewer people are willing to wrestle with long, complex arguments, the ability to do so becomes increasingly scarce. Scarcity breeds value. In a world awash in summaries, videos, and AI-generated explanations, those who continue to read deeply will possess an asymmetric cognitive advantage over those who merely consume other people’s conclusions.

I’ve said for years, if you read one book on a subject, you’re likely the smartest person in the room on that subject because, as Horowitch says, nobody reads books anymore. If you’ve read two books on that same subject, you’re in a set of maybe 500 people in the world who are “The Experts” in the field. You’re not in the top 10, for sure, but you’re in the conversation when somebody asks, “Who should we talk to about this?”

Takeaway

Ironically, Claude Fable articulated the dilemma better than I could:

The more AI floods the world with instant answers, the more the deep reader becomes the rare person who can ask better questions, spot what the summary missed, and tell the difference between knowing about something and actually understanding it.

-- Claude Fable, June 2026

We're living through a rare moment when a clear opportunity arises. We are at the proverbial fork in the road. One path follows convenience (summaries); the other is strategic (book reading). Over time, the people who choose the second path are far more likely to become the experts, the people executives call when the problem is difficult and the stakes are high. That’s the opportunity. That’s the advantage.

So keep using AI. Keep listening to podcasts. I know I intend to do that. But don't let those tools replace the hard work of reading, of reaching for the top levels of Bloom’s Taxonomy.

The old proverb says, “In the land of the blind, the one-eyed man is king.” If Horowitch is right, then in the age of AI, the deep reader becomes the one-eyed king

Source

Rose Horowitch, 2026. The Age of Reading Is Over [Essay]. The Atlantic, URL: https://www.theatlantic.com/magazine/2026/08/reading-crisis-postliterate-age/687618/

References

Alex, 2026. Atlantic Story is absolutely frightening [Comment]. Alex & Books - Substack, URL: https://substack.com/@alexandbooks/note/c-290622997

James Marriott, 2025. The dawn of the post-literate society [Essay]. Cultural Capital - Substack, URL: https://jmarriott.substack.com/p/the-dawn-of-the-post-literate-society-aa1

Nancy Adams, 2015. Bloom’s taxonomy of cognitive learning objectives. [Paper] Journal of Medical Library Association. 103(3):152-3. doi: 10.3163/1536-5050.103.3.010. PMID: 26213509; PMCID: PMC4511057. URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC4511057/

Noah Friedman, Jason Reynolds, 2026. Kids can’t read anymore. Does it matter? [Podcast]. NO SUCH THING, URL: https://www.nosuchthing.show/p/debunking-the-reading-crisis-jason-reynolds

Virginia Clinton-Lisell, 2022. Listening Ears or Reading Eyes: A Meta-Analysis of Reading and Listening Comprehension Comparisons [Academic Paper]. ResearchGate, URL: https://www.researchgate.net/publication/356615999_Listening_Ears_or_Reading_Eyes_A_Meta-Analysis_of_Reading_and_Listening_Comprehension_Comparisons

I’m thoroughly enjoying Matt Dinniman’s series. In my reviews of the first two books, I said that the premise was mind-numbingly stupid; funny and entertaining, with the ridiculousness dial turned all the way to 11, but just completely bonkers. Here’s the premise.

An alien corporation, Borant, destroys every human-made structure on Earth, repurposing the debris into an 18-level underground dungeon spanning the globe. Out of the ~8 billion people living on earth before the attack, roughly 13 million find their way into the dungeon. Carl, our hero, and his girlfriend’s cat, Princess Donut, are two that make it in. By entering, they are forced into a lethal, world-wide D&D game designed for alien entertainment and broadcast across the galaxy. For each floor, the game AI sets a timer for the surviving crawlers to find the staircase to the next floor. When the timer hits zero, the floor collapses with every crawler that didn’t find the staircase. Brutal. Oh, and each floor is filled with D&D-style monsters trying to kill them. It’s The Hunger Games crossed with The Running Man crossed with World of Warcraft.

All of that is still true. But with this third book, I’m changing my mind. It’s not stupid at all. The premise is insane but Dinniman tells it with real craft. The characters are spectacularly unique and have deep emotional arcs. Even the NPCs (think Star Trek Red Shirts) have some depth. And each floor is completely different. There’s none of the rinse-and-repeat I remember from the D&D games of my youth. It’s complicated, but Dinniman is so good at this that the reader is never really lost.

And then there’s Jeff Hays, the narrator. This guy is an amazing actor. The audiobook feels like a full cast. In reality, Hays performs almost all of it himself. Incredible.

And I discovered that Jeff Hays has his own online acting troupe and sound effects business. It’s called Soundbooth Theater. From his website, “Soundbooth Theater is a full-service audio production company that specializes in multicast audiobooks and Cinematic Audio, which are enhanced with sound effects and music.” I listened to the audible version for the first two books. For this one, I tried the Soundbooth version.

I have to say that I prefer the Audible versions. Think of the Soundbooth version as a radio play in the tradition of Orson Welles’ 1938 War of the Worlds broadcast with a full cast, sound effects, and the whole production. But, for my experience, all the extra sounds and fake commercials that Soundbooth added just distracted from the story. I hate to say that because I really want Hays to succeed. He’s such a great talent. Don’t get me wrong. The radio play was fine. I still had a blast and everything. I just prefer the toned down Audible version.

Takeaway

I started this series convinced it was gloriously dumb fun; the literary equivalent of a bag of chips. Three books in, I realize I was wrong about that. Dinniman is doing something real here: building characters you root for inside a world designed to kill them, floor by floor, with no guarantee anyone survives. Carl and Princess Donut are stuck in a dungeon that spans the entire planet, and somehow I’m the one who can’t find the exit. Book 4 is already loaded.

Source

Matt Dinniman (Author), Jeff Hays (Narrator), 2021. The Dungeon Anarchist’s Cookbook (Book 3) [Fiction Book]. Goodreads, URL: https://www.goodreads.com/book/show/211721809-the-dungeon-anarchist-s-cookbook

References

Matt Dinniman (Writer), Jeff Hays, Andrea Parsneau, Ian M. Walker, Justin Thomas James (Narrators), 2026. Dungeon Crawler Carl [Audio Delivery Site]. Soundbooth Theater.
URL: https://soundbooththeater.com/series/dungeon-crawler-carl/

Rick Howard, 2026. Book Review: Dungeon Crawler Carl (Book 1) [Book Review]. First Principles Newsletter - Substack, URL: https://diffuser.substack.com/p/book-review-dungeon-crawler-carl

Rick Howard, 2026. Book Review: Carl’s Doomsday Scenario (Book 2) [Book Review]. First Principles Newsletter - Substack, URL: https://diffuser.substack.com/p/book-review-carls-doomsday-scenario

Staff, 2021. Dungeon Crawler Carl Wiki [Wiki]. Fandom, URL: https://dungeon-crawler-carl.fandom.com/wiki/Dungeon_Crawler_Carl_Wiki

David Webb, 2010. Orson Welles - War Of The Worlds - Complete Radio Broadcast 1938 [Recording]. YouTube, URL:

Note about Using AI for this Essay

I used Claude Code to

• Check for grammar and spelling errors.

• Flag passive voice.

• Suggest changes to awkward sentence construction.

• Create images.

• Fact-check

• Stress-test my thesis

In terms of cybersecurity in the movies and TV, there are really only a handful that managed to get the details right and tell a compelling story. I’ve written about a few of them before. But one of my favorites is Sam Esmail’s Mr. Robot: entertaining TV that shows cyber in a realistic light. I recently stumbled upon a scene from Season 1, Episode 1 where the main character, Elliott, played by Rami Malek, shocks the audience by criticizing one of our sacred cow heroes, Steve Jobs, and claiming that all of our heroes may be counterfeit. It’s a great way to introduce the show.

As America celebrated its 250th anniversary this month, I realized that I’m in the same mindset as Elliott these days. For the first time in many years, the Howard family did not do anything special for the Fourth of July. If you know us at all, that would seem strange, because we usually love this holiday. We are that family: red, white, and blue clothes, flags, bunting, lights, grilled hot dogs, the whole production. We are, in other words, the stereotypical privileged, upper-middle-class white American family. We bought into the country’s origin story hook, line, and sinker.

We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness.

- The Signers of the Declaration of Independence

It was, and is, an exquisite, powerful, and radical vision statement. At the time, most of the world was ruled by kings, queens, emperors, and tyrants. Yet this unlikely collection of people, with little in common beyond occupying the same patch of land, an island of misfit toys really, fought a seemingly hopeless war for the chance to make that vision real. Spoiler Alert! We won. (See the excellent Ken Burns Documentary, The American Revolution, for details. Link below).

Eleven years after the signing of the Declaration, James Madison and others, the Fathers of the Constitution, built the guardrails, the laws, so that these new United States could

form a more perfect Union, establish Justice, insure domestic Tranquility, provide for the common defence, promote the general Welfare, and secure the Blessings of Liberty to ourselves and our Posterity.

- The Framers of the Constitution

I’m reminded of the speech Tom Hanks gives in the vastly underrated Steven Spielberg movie “Bridge of Spies.”

“Your Agent Hoffman right? German Extraction?

My names Donovan, Irish, both sides, mother and father.

I’m Irish, your German, but what makes us both Americans?

Just one thing. One … One, One.

The rule book. We call it the Constitution.

And we agree to the rules. And that’s what makes us Americans.

It’s all that makes us Americans.

It’s all that makes us Americans. I love that line.

Now, the Howards, we’re not complete idiots; crazy for sure, naive absolutely, but not complete idiots. We know this is aspirational. We know it’s not completely true; maybe not even close to being true. It wasn’t back then and it still isn’t today. But we want it to be. We really want it to be

It reminds me of a line from one of our family’s favorite movies, Secondhand Lions. Robert Duvall gives this great speech about what it means to be a man to his young nephew.

Just because something isn’t true doesn’t mean you can’t believe in it.

Sometimes the things that may or may not be true are the things a man needs to believe in the most.

That people are basically good;

that honor, courage, and virtue mean everything;

that power and money, money and power mean nothing;

that good always triumphs over evil;

and I want you to remember this,

that love, true love, never dies.

You remember that, boy.

You remember that.

Doesn’t matter if it’s true or not.

You see, a man should believe in those things, because those are the things worth believing in.

- Second Hand Lions (Tim McCanlies)

So, the Howards are believers.

But there is ample evidence in our collective history that demonstrates that not all of us on this land of misfit toys agree with the vision. The best example of this is the U.S. Civil War. Some 72 years after the ratification of the U.S. Constitution, when the South couldn’t get their collective heads around the idea that all human beings possess the same inherent rights, we fought another unimaginable war in defense of the notion. President Lincoln reminded us all of the reason in his Gettysburg address:

Four score and seven years ago our fathers brought forth, upon this continent, a new nation, conceived in Liberty, and dedicated to the proposition that all men are created equal.

- President Lincoln - Gettysburg Address

Since the beginning, the country has made some tremendous strides to pursue this vision of a “more perfect union;” many times at great cost to the country’s blood and treasure. I’ve collected a highlight reel that demonstrates these best achievements (See A More Perfect Union: The U.S. Highlight Reel of Achievement below). But it’s also made some horrible mistakes setting the country back against its stated vision. I’ve also collected a highlight reel of those mistakes (See A More Perfect Union: One Step Forward, Two Steps Back below). I used ChatGPT to create a chart to show the ups and downs. It’s not perfect, but it gets the idea across.

Takeaway

And that’s the point now I’m trying to make. Today, in conjunction with the 250th anniversary of the country’s founding, we are in a huge downward trough. It seems to us living through it, that it’s the worst it’s ever been, but that simply isn’t true. We’ve been here before. And every single time, the country has found a way to turn itself around again; to reset our goals and aspirations toward that more perfect union. And that’s the sliver of hope I cling to. That whatever we are doing right now as a country, as a culture, as an island of misfit toys, we will find our way again. It’s just not this year.

So, the Howards are not wearing our red, white, and blue accouterments this year. We are not celebrating. We are biding our time and looking for the indicators that our fellow neighbors and our leaders will put us back on the right track.

A More Perfect Union: The U.S. Highlight Reel of Achievement

• 1776: Declaration of Independence

• 1783: The U.S. Revolutionary War officially ends.

• 1788: The Ratification of the U.S. Constitution.

• 1791: Bill of Rights ratified

• 1863: President Lincoln’s Emancipation Proclamation declared enslaved people in rebelling states “forever free.”

• 1866: The U.S. Civil War officially ends.

• 1865–1870: The 13th, 14th, and 15th Amendments abolish slavery, establish birthright citizenship and equal protection, and prohibit race-based voting restrictions, embedding equal citizenship into the Constitution.

• 1913: 17th Amendment: shifted the election of U.S. senators from state legislatures to voters.

• 1920: 19th Amendment: Guaranteed women the constitutional right to vote.

• 1924: Indian Citizenship Act: Granted U.S. citizenship to all Native Americans born in the United States.

• 1941: Executive Order 8802: President Roosevelt banned discriminatory employment practices by federal agencies and war-related companies and unions, and created the Fair Employment Practices Commission.

• 1944–1948: Supreme Court decisions like Smith v. Allwright against white-only primaries, Shelley v. Kraemer against racially restrictive covenants, and Truman’s Executive Order 9981 desegregating the armed forces.

• 1954: Brown v. Board of Education holds that “separate educational facilities are inherently unequal,” dismantling school segregation.

• 1950s–1960s: Mass boycotts, marches, sit-ins, and Freedom Rides force the country to confront segregation and discrimination as incompatible with its stated ideals.

• 1964–1965: Civil Rights Act and Voting Rights Act outlaw discrimination in public accommodations, education, employment, and voting, ending most Jim Crow-era legal barriers.

• 1967: Supreme Court decision to strike down state bans on interracial marriage.

• 1968: Fair Housing Act prohibits discrimination in the sale or rental of housing nationwide.

• 1972: Title IX: Prohibited sex discrimination in federally funded education programs and activities.

• 1988: Civil Liberties Act authorized reparations and a formal presidential apology to Japanese Americans incarcerated during World War II. This is one of the country’s clearest formal acknowledgments of a past constitutional failure.

• 2008–2009: Barack Obama is elected and inaugurated as the first Black president of the United States, a landmark democratic milestone in a country whose Constitution originally tolerated slavery and whose laws later enforced segregation.

• 2015: Supreme Court required states to license and recognize same-sex marriages.

• 2020: Supreme Court holds that firing someone for being gay or transgender violates Title VII’s prohibition on sex discrimination

• 2022: Respect for Marriage Act: once a marriage is real in one place, the government has to treat it as real everywhere.

A More Perfect Union: One Step Forward, Two Steps Back

• 1787–1865: Chattel slavery is legal and central to the U.S. economy and politics

• 1798: Alien and Sedition Acts: The new republic quickly violated its own speech and liberty principles by restricting foreign-born residents and criminalizing criticism of the federal government.

• 1830s: Indian Removal Act and Trail of Tears: Federal policy used treaties, coercion, and forced displacement to remove Native nations from ancestral lands, often violating treaty commitments and even Supreme Court rulings.

• 1857 – Dred Scott v. Sandford: Supreme Court rules Black people cannot be U.S. citizens and are not protected by the Constitution.

• 1850: Fugitive Slave ActMade the federal government an active enforcement arm of slavery by requiring escaped enslaved people to be returned and penalizing those who helped them.

• 1865–1877: Black Codes and the violent overthrow of Reconstruction: Southern states used law, terror, and economic coercion to recreate slavery-like conditions after emancipation. This is the essential bridge between the 13th Amendment and Jim Crow.

• 1873: The Supreme Court sharply narrowed the Privileges or Immunities Clause of the 14th Amendment, weakening one of Reconstruction’s intended protections for national citizenship.

• 1876: The Supreme Court limited federal power to prosecute private racial violence, making it harder for the federal government to protect Black citizens from white supremacist terror.

• 1877: End of Reconstruction: Federal retreat from Reconstruction allowed white supremacist state governments to regain control and dismantle Black political power across the South.

• 1883: The Supreme Court struck down key parts of the Civil Rights Act of 1875, leaving private discrimination in inns, theaters, railroads, and other public accommodations largely beyond federal reach

• 1887: Federal allotment policy broke up tribal landholding, weakened Native sovereignty, and transferred massive amounts of Native land out of Native control

• 1896: Plessy v. Ferguson: Constitutionalized “separate but equal” and gave federal blessing to racial segregation for decades.

• 1898: Wilmington coup and massacreWhite supremacists violently overthrew a biracial local government in Wilmington, North Carolina. It is one of the clearest examples of anti-democratic racial violence in U.S. history.

• 1910s–1940s: Racial terror lynching and anti-Black massacres: Tulsa, East St. Louis, Elaine, Rosewood, and other episodes show that private violence and public complicity worked together to enforce racial hierarchy. The Smithsonian describes Tulsa’s 1921 destruction of Greenwood as one of the deadliest racial massacres in U.S. history.

• 1920s–1970s: Local and federal government instituted Indian boarding schools and forced assimilation by removing Native children from families and communities to suppress Indigenous languages, religions, and cultures.

• 1930s–1960s: Federal redlining and housing exclusion helped produce racially segregated neighborhoods and long-term wealth gaps.

• 1942–1945: President Roosevelt signed Executive Order 9066 to authorize the incarceration of Japanese Americans during World War II.

  20th century: Discriminatory housing policies, job discrimination, and union practices create racially segregated neighborhoods and persistent economic gaps.

• 1950s–1970s: Federal intelligence and law-enforcement agencies surveilled and disrupted civil-rights, Black liberation (including Dr. Martin Luther King, antiwar, Native, and other domestic political movements.

• Post–1960s: Despite civil rights laws, failures to enforce protections and resistance to integration slow or stall equality gains.

• 1996: Defense of Marriage Act denied federal recognition to same-sex marriages and allowed states to refuse recognition of same-sex marriages performed elsewhere, writing anti-LGBTQ exclusion into federal law until later court decisions and repeal.

• 21st century: Ongoing disparities in education, income, wealth, and exposure to criminal justice and incarceration between racial groups.

• President Trump’s Second Term

  • Implementation of ideas from Project 2025 and related agendas to shrink or repurpose civil rights enforcement like limiting cutting civil rights data collection and targeting DEI offices and programs.

  • Supreme Court decisions and federal posture that weakens the Voting Rights Act’s ability to protect minority voters.

  • Rolling back protections for women and LGBTQ+ people.

  • The removal of senior military leaders, including women and Black officers, raised concerns that ideology and anti-DEI politics were influencing decisions traditionally grounded in merit, experience, and operational need.

  • Public office used for family enrichment, patronage, and selective impunity corrodes the equal-citizenship principle by making access to law, opportunity, and protection depend on proximity to power rather than equal standing under the Constitution.

  • Using the justice system to punish political opponents turns equal justice into a loyalty test, where power decides who gets protected and who gets prosecuted

  • False or exaggerated claims of voter fraud turn democracy into a suspicion machine, where rare irregularities are used to justify broad restrictions that make it harder for eligible people to vote

  • 2022: The Supreme Court overruled Roe and Casey, eliminating the federal constitutional right to abortion and returning abortion regulation to the states.

  • 2024: The Supreme Court held that presidents have immunity for official acts, expanding constitutional protection around presidential power and raising concerns about accountability for abuses of office.

  • 2025: President Trump revoked Executive Order 11246’ essentially ending DEI programs and The administration framed this as restoring merit and colorblind equality; critics framed it as dismantling civil-rights infrastructure.

Source

Sam Esmail (Creator), Rami Malek, Christian Slater, Carly Chaikin (Actors), 2015 - 2019. Mr. Robot [Wiki Entry]. IMDB - YouTube, URL:

References

Abraham Lincoln, 1863. The Gettysburg Address [Historical Document]. Abraham Lincoln Online, URL: https://www.abrahamlincolnonline.org/lincoln/speeches/gettysburg.htm

Alan Jenkins, 2018. Civil Rights Today [Essay]. American Experience | PBS, URL: https://www.pbs.org/wgbh/americanexperience/features/eyesontheprize-civil-rights-today/

Framers of the Constitution, 1787. The Constitution of the United States: A Transcription [Historical Document]. National Archives, URL: https://www.archives.gov/founding-docs/constitution-transcript

Frank W. Munger, Carroll Seron, 2017. Race, Law, and Inequality, 50 Years After the Civil Rights Era [journalArticle]. Annual Review of Law and Social Science, URL: https://digitalcommons.nyls.edu/cgi/viewcontent.cgi?article=2370&context=fac_articles_chapters

Geoffrey C. Ward (Writer), Ken Burns, David Schmidt Sarah Botstein, and Buddy Squires (Directors), 2025. The American Revolution [Wiki Entry]. Letterboxd, URL: https://letterboxd.com/film/the-american-revolution-2025/

Lukas Althoff, Hannah Reichardt, 2024. Why history continues to shape racial inequality in the US [History]. VoxEU / CEPR, URL: https://cepr.org/voxeu/columns/why-history-continues-shape-racial-inequality-us

Rob Reiner (Director), Aaron Sorkin (Writer), Michael Douglas (Actor), 1995. The American President [Movie]. Letterboxd. URL https://letterboxd.com/film/the-american-president/

Staff, N.D. Civil Rights Era - Timeline [Wiki Entry]. Jim Crow Museum - Ferris State University, URL: https://jimcrowmuseum.ferris.edu/timeline/civilrights.htm

Staff, 2018. From Slavery to Segregation [Website Article]. Equal Justice Initiative, URL: https://segregationinamerica.eji.org/report/from-slavery-to-segregation.html

Staff, N.D.. U.S. History Primary Source Timeline: The Civil Rights Movement [Wiki Entry]. Library of Congress, URL: https://www.loc.gov/classroom-materials/united-states-history-primary-source-timeline/post-war-united-states-1945-1968/civil-rights-movement/

Staff, N.D. Reconstruction and Jim Crow Eras - A Brief History of Civil Rights in the United States [Wiki Entry]. HUSL Library, Howard University School of Law, URL: https://library.law.howard.edu/civilrightshistory/blackrights/jimcrow

Staff, N.D. Why Treaties Matter: Broken Promises [Wiki Entry]. Smithsonian National Museum of the American Indian, Indian Affairs Council of Minnesota, Minnesota Humanities Center, URL: https://treatiesmatter.org/exhibit/welcome/broken-promises/

Steven Spielberg (Director), Tom Hanks (Actor), 2015. Bridge of Spies [Movie]. Letterboxd. URL https://letterboxd.com/film/bridge-of-spies

Thomas Jefferson (Original Author) and the Signers of the Declaration of Independence, 1776. The Declaration of Independence [Historical Document]. National Archives, URL: https://www.archives.gov/founding-docs/declaration-transcript

Willoughby Alley, 2016. What Makes Us Americans [WWW Document]. Bridge Of Spies Clip - YouTube. URL

Lee Hoedl, 2020. The American President - The Final Speech [Movie Clip]. YouTube. URL

Tim McCanlies (Writer / Director), Robert Duvall and Haley Joel Osment (Actors). 2003. The Speech: Secondhand Lions. [Movie] Youtube. URL: