Kayfable
What does Anthropic actually want?
Hey there, I’m Alberto! 👋 Each week, I publish long-form AI analysis covering culture, philosophy, and business for The Algorithmic Bridge. Paid subscribers also get Monday how-to guides and Friday news commentary. I publish occasional extra articles. If you’d like to become a paid subscriber, here’s a button for that:
Interrupting the regular schedule because the Anthropic-Fable-US government situation demands it. A recounting of the facts as known, plus my analysis of what's really going on. Monday's practical guide moves to tomorrow (Tuesday).
kayfable noun
kay·fa·ble | \ ˈkā-ˌfā-bəl \
1: a fictitious story whose lesson depends on everyone pretending it is real
2: a public performance maintained past the point of plausibility
3: whatever shit show is going on between Anthropic and the US government
So let’s start with the facts as reported by the press.
April 7th. Anthropic announces Mythos Preview, a model with unprecedented coding and agentic capabilities. They don’t release it, alleging serious risk.
June 9th. Anthropic releases Fable 5, a Mythos-class model and the best in the world. It’s capped by guardrails; prompts related to cybersecurity, biology, distillation, and AI research are routed to an inferior model. People are awed and angry in equal measure.
June 11th. Amazon CEO contacts the US government (USG) with info on a jailbreak that reveals Fable 5 could be really dangerous. The USG reportedly presses Anthropic to fix or shut down the model, over concerns that China could access the model (Anthropic denies the USG ever said anything about China). Anthropic refuses, allegedly stating that the presumed danger is a known issue and has been exaggerated. Surprised by Anthropic’s reluctance, the USG puts Fable under the export control directive, forcing Anthropic to stop serving Fable to any foreign national, including people working at Anthropic.
June 12th. Anthropic retires Fable for all customers “to ensure compliance.”
June 14th. Anthropic staffers fly to Washington to fix the situation.
June 15th. They’ve met with the Trump administration.
At the time of publishing, nothing else has happened.
Among the long chain of implausible events that have surrounded Anthropic in 2026—including but not limited to a wild revenue surge, OpenAI *sorpasso,* and **repeated clashes with the USG—there’s one question that stands out above every other: why not just obey?
As the poster child AI safety company, why not try to solve the supposed jailbreak? Why not concede that the problem—even if it’s a negligible issue to you—is a real threat? Why not give the USG what they ask for and the public a coherent story of who you are? Why force the USG’s hand to put Fable 5 under an unfair regulatory stronghold? Why, when you’re familiar with the USG doing crazy stuff like designing a domestic company a “supply chain risk,” don’t you just shut up?
In Anthropic’s defense, I will say that patching a jailbreak is not trivial—something else will break from jail tomorrow—so I didn’t expect a trivial answer. A refusal was fair. But Anthropic could have refused on the grounds of not allowing a rival—even if one that’s also an investor, as Amazon is—to set a precedent by which the USG can shut down a released product due to a competitor’s panic call. That’s a rather autocratic thing to do, more so when the problem exists in products by other companies. But, as far as we know, Anthropic chose a different line of argument: this is a minor issue. The reason, as I see it, is this: if Anthropic had quietly accepted, fixed, and re-released Fable a couple of weeks later, they wouldn’t have any martyrdom to feed from. They would give up precious aura points. As things stand, Anthropic gains three things they wouldn’t have had they obliged:
They remain the underdog, smaller and less powerful than both OpenAI and Google, beaten and re-beaten by the USG; everyone tries to prevent them from being the AI company of The People.
They reinforce the safety thesis: we said Mythos was dangerous and worked hard to make it safe, but look, even the USG asks us to retire our models due to AI-related dangers.
Compute is scarce and not serving Fable stops the cash hemorrhage.
Setting aside the terrible decision-making of the USG, it’s pretty clear that Anthropic is making this situation worse than it could be for its own gain. I respect Anthropic as a tech company—and I can see, like everyone else, that the USG has its sights on them—but I’m tired of the theater. They’re courting Trump at court while telling us with their actions how mistreated they are. And now this kayfable stops me from using Fable. I’m sure they want it resolved quickly. But I’m also confident they’re not completely unhappy that it happened. This incident is a piece of a much larger puzzle, and when you see the full picture, you will realize that things are still going according to plan.
You have to understand that Anthropic’s ultimate goal is not to build AGI, but to own it. To control it. They don’t trust you or me with it, and never have. Their safety stance has never been an “AI is unsafe” stance, but an “AI is unsafe in your unsafe hands” stance. (This is why Sam Altman—always fickle, always reactive—has pivoted OpenAI to be the “personal AGI for everyone” company, to overtly oppose Anthropic’s stance.) Let’s take a look at Anthropic’s recent behavior.
They held back Mythos Preview but did so quite publicly. Do you think that was the first time a company held back an AI model? That’s the standard modus operandi; they never share with us the best they have. Why make a fuss over Mythos? You guessed it: to get the USG’s attention.
They’ve insisted on chip export controls for months while comparing it to giving nukes to North Korea. These kinds of comparisons are the equivalent of shouting to the USG: “Take me! Take me!” You wouldn’t let your neighbor have the launch codes, would you?
They repeat ad nauseam that we are all going to lose our jobs. The jobpocalypse, a white-collar bloodbath, etc. Amodei gives it 5 years for 20% of the office economy to be replaced by agent swarms. Who can stop this? You guessed it: the USG. Be it UBI or a job guarantee or an influx of investment in the relational sector—only the USG can fix it.
They only want “a few trusted partners” for their projects and for the future of AI, etc. A coordinated coalition of the kind reminiscent of “everything for the people without the people”—or worse, an oligopoly.
The latest talking point is recursive self-improvement: “By the end of 2028, AI will be able to make itself better without any sort of human involvement. No, wait—OpenAI says it will happen by March 2028? Then we say by late 2027! USG—come save us, no?”
Anthropic is a deeply undemocratic company in the sense that it has never been their plan to give us access to this technology; to them, the distribution of AI products is an unwanted by-product of, on the one hand, needing both our money and our data, and, on the other hand, of the damned AI race sparked by ChatGPT in 2022. They never wanted any of this. Pausing or slowing down AI has never been about R&D but about deployment. The tech is fine, it’s just not so fine if you have it. Amodei distrusts Xi Jinping so much—more than anyone else in Silicon Valley—precisely because his company is the one that most closely resembles China.
Anthropic is working against the clock, but not to build better models like everyone else is. A victory in that front will come naturally to them. They spend 7 hours of every 8-hour workday talking philosophy and safety shenanigans, and merely the remaining hour is dedicated to beating OpenAI and Google on benchmarks, all the while torturing the stock market by disrupting entire SaaS markets with every single tweet. Talent moves around in the Bay Area, yes, but talent doesn’t leave Anthropic. It is the most powerful attractor, like a black hole: for AI guys, Anthropic is the killer or FOMO, the end of elsewhere. Every single co-founder is still there, unlike essentially every other AI lab that matters. Andrej Karpathy is there. Ilya Sutskever will end there (speculation alert). They are, at their core, principled people. They’re working against the clock, but making models is the least of their worries; instead, they want to 1) convince the world that their principles are the principles to be had, and 2) ensure that if we won’t have them, then they’ll be imposed on us. How do they plan to do that? By “tricking” the USG to prioritize the safety story, which is the control story, which is, as I’m telling you, the AI as an undemocratic technology story. The news on Fable suggests it’s a done job.
Will Fable 5 be re-released? Yes, sure. Someday.
Will Anthropic keep releasing models forever as they have before? No. Hard no.
“But Alberto, they can’t work out their huge financial needs (spending commitments, fundraising, IPO, etc.) if they don’t make sure the rest are forced to do the same.” Yes, as long as the industry remains commercial, they’re trapped in a race they’ve already won. But if at some point the USG transforms the industry into a national AI initiative akin to the Manhattan Project, then Anthropic will let go of the facade. They won’t need us—consumers and enterprises alike—even for appearances. Our data is theirs. Our money, inconsequential. As soon as that happens—and I contend that it will happen soon-ish—Anthropic will immediately cut off its technology from the world. It won’t be available even in closed form. “It’s not a toy,” they will say, “and you’re just a bunch of kids.”
If you are still unconvinced by my arguments, let me ask you this in all seriousness: What did you think getting closer to AGI meant?
Vibes? Papers? Essays? Models accessible to all? Did you really think they’d give a superhuman-level AI to everyone? The most important technology humanity has ever invented, in your commoners’ hands? A “country of geniuses in a datacenter” in your backyard? Please, don’t be naive. If you thought that, let me tell you with no acrimony that you’re either too vague in your beliefs to understand what AGI actually is, too innocent in a selfish world, or a fearful atheist: If nothing else, you display a profound lack of conviction on what AI can be; the machine god won’t be pleased. If you were as convinced as the Anthropic guys are, you would have already accepted that you will never set your sights on The Final Thing. You can be unconvinced about AI’s potential—that’s internally coherent—but know these guys are as convinced as it’s humanly possible to be. Whether you get Magnum Opus-10 in your hands depends on their beliefs, not yours.
When the Anthropic guys are called a religious cult, I never take it as a derogatory insult. For a person who has thought through the implications of having a superintelligence live among us, “cultist” is purely definitional. I recently wrote that, except Apple, all the other tech companies are “fearful atheists”—just like you—only paying homage to the machine god in seldom outbursts of terror, but never out of faith. Apple is, in its calm demeanor toward AI, the only true non-believer, oozing defiant confidence. But I made an exception of Anthropic because it is as coherent in its conviction as Apple, but with opposite valence: a company of loyal devotees.
That’s why they give Claude a personality, a blog, and a constitution; that’s why they’re not afraid to say it has emotions, introspection, personhood, and could be conscious; that’s why they advocate for robot rights and model welfare. If you think this is still just marketing, you have understood nothing, and your lack of imagination baffles me more than Anthropic’s behavior of late. There’s no need for marketing when you believe to be sailing toward the end of time. What you need is to be at the helm. Or, failing that, convince those who are to take your preferred course. No amount of shameless kayfable is too much to ensure the world steers itself their way.