A national-security directive forced Anthropic to abruptly cut off Fable 5 and Mythos 5 for all foreign nationals worldwide — the first time the US has hit the kill switch on a frontier model. The pretext is a thin jailbreak; the real story is a precedent that splits the world by model access and turns Anthropic's safety rhetoric into the rope it hangs by.
The Fable shutdown wasn't about a jailbreak — it was a kill switch, and Anthropic spent a year loading it.
The US government invoked national-security export controls to force Anthropic to disable Fable 5 and Mythos 5 for every foreign national on earth — including its own non-citizen employees — over a narrow jailbreak that surfaced already-public vulnerabilities. The industry's fury splits two ways at once: the government's pretext is cartoonishly thin and its export strategy incoherent, and Anthropic spent a year insisting its models were too dangerous for anyone but itself to be trusted with — then acted shocked when Washington took that premise literally. Either way, the precedent is the story: model access is now a sovereign weapon, and no lab's deployment is safe from a 5 PM Friday directive.
Citing national security authorities, the government issued an export control directive suspending all access to Fable 5 and Mythos 5 by any foreign national, inside or outside the US — including Anthropic's own foreign-national employees. The net effect: Anthropic had to abruptly disable both models for all customers. Other Claude models are unaffected.
— Anthropic, in its statement on the suspension. Anthropic's public statement framed the shutdown as an error rather than a justified safety action, apologizing to customers while pledging to get the models back online.
The Wall Street Journal reported Commerce Secretary Howard Lutnick sent CEO Dario Amodei a letter declaring Fable 5 and Mythos 5 subject to export restrictions — barring use by customers outside the US and by foreign nationals within it.
Anthropic says the directive arrived at 5:21 PM with no specific details, but rests on a method of bypassing Fable 5. On review, the technique surfaced only a few previously-known minor vulnerabilities that other publicly available models find too — essentially asking the model to read a codebase and fix software flaws.
— Anthropic, in its blog post on the directive. Anthropic argued perfect jailbreak resistance is impossible for any provider, so recalling a model deployed to hundreds of millions over a narrow potential jailbreak sets an unworkable bar for the entire industry.
The WSJ reported the research was done by Amazon researchers — an Anthropic investor and Project Glasswing partner — who used a prompt series to extract info on a handful of known security vulnerabilities. Notably, the Journal did not report that Amazon shared the findings with the government.
Per Prinz's timeline assembled from Axios reporting, the government contacted Anthropic to ask it to pause releasing the models and was unsuccessful — a refusal many read as setting up the confrontation that followed.
A large share of Anthropic's technical staff — including names like Andrej Karpathy — are not US citizens but here on visas such as EB-1. Under the directive, those employees are barred from interacting with Fable 5 and Mythos 5 internally.
— Dan Robustus, on X. Dan Robustus on X captured the whole tenor of the discourse — a fight in which almost no party emerges looking good.
— Bindu Reddy. AI entrepreneur Bindu Reddy called the move 'really stupid,' noting any model can be coaxed into discussing common security vulnerabilities — and the cluelessness of the government, she said, is astounding.
— Dean Ball. AI policy expert Dean Ball voiced a widely shared confusion about the directive's actual motive — and a shared verdict on its competence.
— Chris Maguire, Council on Foreign Relations. CFR's Chris Maguire backs targeted model export controls but blasted Commerce and BIS for an 'incoherent and self-defeating' strategy — sending chips to China while blocking US companies from releasing their own models.
Just weeks earlier, the White House OSTP account had pushed back on reporting about AI oversight, insisting it was not conducting oversight of all new models — because that 'level of government overreach would have chilling effects on free speech and innovation.' Critics resurfaced it as direct hypocrisy.
— Kirsten Davies, Department of War CIO. Department of War CIO Kirsten Davies posted in support of the action — a pointed shot at Anthropic that, to many, signaled the dispute is about the government's relationship with the company, not Fable 5 itself.
— Sarah Hooker. AI builder Sarah Hooker captured the industry's scorn: the posture that everyone should be grateful to touch an intentionally hobbled technology — and that no one else should be allowed to build it because it's too dangerous.
Critics resurfaced an Anthropic blog line arguing the government 'should have the power to block or deter deployment' of risky models. Will Manidis summarized the whiplash: Dario 48 hours ago said the government should be able to block deployments — and now that it has, says 'not like that.' Anthropic's caveat: it wanted a transparent, fair statutory process, which this wasn't.
— Nick Carter. Investor Nick Carter's sarcasm summed up a dominant view: Anthropic's relentless danger framing stoked the very regulatory panic that just took its flagship models offline.
— Banteg, on X. Banteg's one-liner became the episode's most concise indictment of Anthropic's self-inflicted predicament.
— Aaron Levie. Aaron Levie called it a major turning point for AI regulation — and argued we're unlikely to return to a world where the government isn't far more involved in the pace of AI progress.
Brian Zhao argued restoring access may require ID/citizenship verification not just on Claude but everywhere Fable is served — Cursor, Devin, OpenRouter, a law firm on Harvey. He also noted OpenAI and Google now have little incentive to ship anything Mythos-caliber, since any jailbreak could trigger the same export controls.
NLW compared it to when Sam Altman was ousted and reinstated at OpenAI — the moment Microsoft began quietly building resilience outside OpenAI's models. Once a partner proves itself too capricious to trust, the relationship is permanently reshaped, even if it nominally continues.
— Connor Brown. Connor Brown framed the moment as the opening of the AI wars — KYC and 'anti-compute laundering' laws for frontier models — and asked whether the public will ever have access to frontier intelligence again.
NLW argued the American economy effectively depends on Anthropic and OpenAI revenue rising and investors continuing to fund the buildout. The damage from this move, he said, isn't just to Anthropic — it's hard to overstate the hit to the entire economy.
Commentators argued the shutdown is a warning shot for middle powers: frontier access is not guaranteed. Gale Weiner noted the US narrative advantage over China — predictable, rule-of-law provider versus arbitrary actor — just evaporated, handing procurement officers worldwide a defensible case for sovereign AI and even Chinese open-weight alternatives.
— Mal, on X. Mal on X warned the government is creating a caste system based on access to intelligence: a divide not between rich and poor but between those allowed to think at the frontier and those who simply happen to be citizens of another country.