ANTHROPIC CUTS ACCESS TO FABLE, GOVT. DIRECTIVE

The US government ordered Anthropic to shut off access to its two most advanced models, Fable 5 and Mythos 5, for any foreign national, including the company's own employees. With no way to implement that restriction selectively, Anthropic pulled access for everyone. The directive arrived shortly after Amazon CEO Andy Jassy raised security concerns with the White House based on internal research.

Fable 5 had been in the wild for just days. It promised a significant step up in coding, reasoning, and agentic actions, and Anthropic had spent months working with authorities in the US and the UK, as well as multiple third-party agencies and internal teams, to get it safe to deploy. The company described its safeguards as "substantially more effective than those of any previously deployed model". None of that mattered once the export control directive landed at 5.21 pm Eastern time.

THE RESEARCH THAT TRIGGERED IT

The chain of events started with a paper from Amazon claiming it had, through a series of carefully crafted prompts, gotten Fable 5 to serve up information useful in cyberattacks. Jassy shared the findings with the White House, and the government made the call to block the models to foreign nationals.

Anthropic pushed back hard. It argued the vulnerability was narrow and non-universal, and that similar outputs could be elicited from other frontier models, including GPT 5.5. It disputed the government's characterization of the issue as a "jailbreak". Katie Moussouris reviewed the work and was blunt: "I’ve seen the paper. It’s not a jailbreak.".

The question of what constitutes a jailbreak matters, especially If a model can be prompted to produce harmful information through a specific chain of inputs, that is a bug the developer needs to fix. But if the model is basically operating as designed and a creative user can get it to answer questions it was trained to avoid, the line gets fuzzier.

THE COLLATERAL DAMAGE

Because the directive applied to any foreign national, both inside and outside the United States, Anthropic's own researchers were barred from accessing the models they helped build. Many of the company's researchers are foreign-born. The company said it had no technical way to apply the rule selectively, so it had to turn off access for everyone.

The result: paying customers lost access to the products they had been using. Anthropic apologized publicly, calling it "We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible." and reiterating its view that the vulnerability was "narrow" and "non-universal". It warned that "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers," if the standard were applied across the industry, which would halt all new model deployments for all frontier model providers.

Users were not charitable, some called it an "absolute nightmare scenario" and another said "this has to be a joke". A company that had been positioning itself as the responsible steward of safe AI development was now blocking its own employees and customers because of a vulnerability it disputes.

THE POLITICS MAKE IT WORSE

Anthropic and the Trump administration have been at odds for some time. In February, Trump ordered federal agencies to stop using Anthropic's AI. Defense Secretary Pete Hegseth designated the company a supply chain risk. The backdrop: Anthropic has refused to allow its AI to be used for mass surveillance of Americans or to power lethal autonomous weapons.

The two sides seemed to have made amends recently, working together to expand access to Mythos. Now they are destined to clash again. A former Commerce Department official, Kate Koren, speculated to the Wall Street Journal that the White House's dislike of Anthropic may have played a role in the decision. The administration's own research arm appears to have been the trigger, via Amazon's Jassy, but the speed and breadth of the response suggest more than technical concern.

WHAT HAPPENS NOW

Anthropic says it is working to restore access as soon as possible. But the path is unclear. If the government insists the vulnerability is real and demands a permanent fix, Anthropic will have to decide whether to comply or fight publicly. A fight would be risky: the startup depends on US infrastructure and talent, and the government has shown it can act fast.

The broader industry implication is larger than this one dispute. The precedent of a preemptive, company-wide export control based on a contested research finding, applied to employees as well as external users, is something every AI developer will be watching. If the standard sticks, no frontier model will launch without government preapproval of who can use it. That is not a hypothetical future. It is the present for Anthropic's two most important products.

For now, customers are locked out, researchers are locked out, and the company is operating on a directive it does not agree with. The model that the government says is dangerous remains blocked. The model that Anthropic says is the safest it has ever shipped remains blocked. And neither side seems willing to admit the line between a vulnerability and a policy preference has been crossed.