The Anthropic Fable Mythos takedown unfolded with startling speed, after the Trump administration reportedly gave the AI company just 90 minutes on Friday to pull down its most powerful models before slapping it with a licensing regime. According to an Anthropic source, what followed was a tense scramble that ended with the company abruptly disabling its flagship systems for every customer worldwide.

A Sudden Ultimatum

The episode began with a phone call. At 1 p.m. ET on Friday, Anthropic was contacted by the government and instructed to roll back the release of its Mythos and Fable models, citing a “national security threat,” but offering no further details, the source said.

The demand caught the company off guard for a specific reason. The government and Anthropic had already collaborated on pre-release testing of Fable, and the company had received explicit approval to deploy the model. Now, that approval was effectively being reversed in real time.

Anthropic’s first instinct was to fix whatever was wrong. As the source put it, the company immediately sought to understand the specific nature of the threat so it could remediate it. The government, however, held firm on its demand rather than detailing the problem.

The Commerce Department Letter

The pressure escalated through the afternoon. By around 5:30 p.m. ET, Anthropic received a formal letter from the Commerce Department imposing controls on where the models could be used and by whom.

The scope of that order was sweeping. According to the company, it was instructed to suspend all access to the Fable 5 and Mythos 5 models by any foreign national, whether inside or outside the United States, including Anthropic’s own non-citizen employees. Faced with such a broad directive, the company concluded it had no choice but to cut off access to the models for all customers in order to ensure compliance, and immediately began de-deploying them.

Crucially, the restriction applied only to the two most powerful systems. Access to Anthropic’s other Claude models, including its latest Claude Opus 4.8, was not affected.

What Triggered the Crackdown

The drama reflects a frantic effort within the government to respond to alarm over the models’ capabilities. Reporting indicates the action followed warnings, including from rival companies, about what the technology could do.

The central concern appears to have been cybersecurity. The newly launched Fable 5 was described as a Mythos-class model, a tier Anthropic positioned above its previous Opus-class systems, and it was said to be particularly effective at identifying software vulnerabilities. That very strength raised fears it could be exploited in cyberattacks.

According to administration accounts, the decisive moment came when another company claimed it had managed to jailbreak Mythos, getting around its internal safety guardrails. That demonstration alarmed officials about possible national security risks and, after an unsuccessful attempt to get Anthropic to pause the release, prompted the export control letter.

A Dispute Over the Threat

Anthropic has pushed back on the characterization of the danger. The company said the government did not provide specific details about its national security concern in the letter itself.

It did, however, acknowledge becoming aware of the jailbreaking technique at the center of the dispute. The company said it reviewed a demonstration of the method being used to identify a small number of previously known, minor vulnerabilities, describing them as relatively simple flaws that other models could also discover. In its public response, Anthropic framed the situation as a misunderstanding and said it was working to restore access as soon as possible, while apologizing to customers for the disruption.

The Broader Context

The standoff did not emerge from nowhere. Earlier in the month, the Trump administration had issued an executive order requiring the most advanced AI models to be tested before deployment, and Anthropic maintains a partnership with a testing body housed at the Commerce Department.

The Mythos line had already drawn intense attention. The system, designed to find software vulnerabilities, had captivated both Wall Street and government officials with its cybersecurity capabilities, and Anthropic had limited its broader rollout to a select group of companies through a cybersecurity initiative known as Project Glasswing. From the government’s perspective, one official suggested the models needed to remain locked down until the national security apparatus was hardened, a step expected to take a matter of weeks.

What’s at Stake

The implications stretch well beyond a single weekend of disruption. Anthropic has argued that the standard being applied could effectively halt all new frontier model deployments across the AI industry, raising broad questions about how cutting-edge systems can be released at all.

There are commercial stakes as well. The company recently moved toward a public listing, and the export control decision could give investors pause, prompting concern about whether Anthropic can stay at the frontier of AI development if the government continues to single out its models for restrictions.

The Bigger Picture

The forced takedown represents one of the furthest-reaching actions the government has taken in response to the capabilities of an AI model. It captures a central tension of the current moment: the same features that make a model impressive, in this case the ability to find software flaws, are precisely what make regulators nervous.

For now, the two Mythos-class models remain offline as Anthropic works to address the government’s concerns and restore access. How quickly that happens, and on what terms, may help define the emerging rules of the road for releasing the most powerful AI systems, and how much say the government will have over when, and to whom, they can be deployed.