Anthropic’s security warnings may have just backfired – the government has pulled the plug on its super-powered AI.

The US government on Friday ordered Anthropic to immediately block access to two of its most powerful AI models – Claude Fable 5 and Claude Mythos 5 – citing national security concerns. Anthropic announced to X that it complied, but made it clear that it thought the government had made a mistake.

The order, which Anthropic said it received Friday at 5:21 p.m. ET, forces the company to disable both models for all users worldwide — not just foreigners the government’s export control order was nominally aimed at. Access to other Anthropic models is not affected.

Why does any of this matter? Mythos is Anthropic’s most capable AI model, which the company previewed in early April and has been tightly limited ever since due to what Anthropic described as its unique ability to detect security vulnerabilities in software. According to Anthropic, Mythos identified flaws in every major application and web browser it tested, so instead of releasing it widely, the company launched a controlled program called Project Glasswing, sharing it with about 50 test organizations, including Amazon, Apple, Google, Microsoft, and CrowdStrike, to use for cybersecurity defense work.

Mythos 5, released just three days ago, was Anthropic’s response to obvious commercial pressure: a version of Mythos equipped with security controls that block responses in high-risk areas such as cybersecurity and biology, making it safe enough for general release, the company argued. It quickly became the most powerful AI model available to the public, according to benchmark tests from Vals AI, a company that tracks the performance of AI technology.

The government order was put in place as an export control measure, preventing foreign access to the models. But in a lengthy blog post, Anthropic says its understanding is that the primary concern is a jailbreak for Fable 5. So far, the company says, the government has provided only verbal evidence of “potentially complex, non-universal jailbreaks” – which, as Anthropic explains, amounts to prompting the model to read a specific codebase and identify software errors. And by the way, adds the company, it’s a “level of skill” that’s already widely available in other publicly accessible models, including OpenAI’s GPT-5.5. It is also regularly used by cybersecurity professionals for defensive purposes, Anthropic said.

Anthropic’s broadest argument is that its strongest defenses are powered by independent classification systems that operate separately from the model itself, meaning that even if someone convinces Fable to keep talking past rejection, the basic defenses against the most dangerous effects are still there.

Obviously, none of this is enough to stop the government from acting, and Anthropic doesn’t hide its frustration. “We do not agree that the discovery of a potential prison massacre should be the cause of a recall of a commercial model that has been shipped to hundreds of millions of people,” the company wrote. “If this standard were implemented across the industry, we believe it would stop new model shipments from all frontier model suppliers.”

Anthropic is widely expected to pursue an IPO this year and play a major role in its public image of being a more safety-conscious alternative to its competitors. What’s ironic to onlookers is that the very warning Anthropic expressed in restricting the Mythos – which it proposed as a model too dangerous to release publicly – has now apparently attracted the kind of government scrutiny that could seriously disrupt its business.

OpenAI’s Sam Altman must be enjoying this, at least. In April, he told host Ashlee Vance that Anthropic’s handling of the Mythos amounts to “fear-based marketing.” “It’s an amazing marketing thing to say, ‘We built a bomb. We were going to drop it on your head. We’re going to sell you a bomb shelter for $100 million,'” Altman said. Altman, whose company is also expected to pursue an IPO as soon as possible, did not predict a government shutdown, but pointed to something that has come back to bite Anthropic in the meantime, which is that if you spend months telling the world your AI is a unique threat, the world — including the US government — tends to listen.