The US government has ordered Anthropic to suspend access to its two most capable AI models, Fable 5 and Mythos 5, in what appears to be the first time Washington has forced a commercial AI product offline. The directive, which Anthropic says it received at 5:21pm ET on 12 June, cites national security authorities and demands the company cut off all foreign nationals, including its own employees.

Because Anthropic cannot reliably distinguish foreign nationals from domestic users in real time, the practical result is a hard shutoff of both models for every customer worldwide. Access to all of Anthropic’s other models remains unaffected.

Three days from launch to shutdown

Anthropic launched Fable 5 on 9 June as its first “Mythos-class” model available to the general public. Fable 5 is the public-facing version, with safety classifiers that route flagged requests to the weaker Claude Opus 4.8, while Mythos 5, the same underlying model with cyber safeguards lifted, was restricted to vetted cybersecurity defenders and critical infrastructure operators.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Both models were already in the crosshairs of the Trump administration, which had previously moved to block Anthropic’s broader Mythos rollout. The company had offered Fable 5 free to Pro, Max, Team, and Enterprise subscribers through 22 June.

The jailbreak dispute

The Commerce Department reportedly acted after an unnamed company claimed it had found a way to jailbreak Mythos, according to Axios. Anthropic says it has only received verbal evidence of a narrow, non-universal jailbreak that essentially involves asking the model to read a specific codebase and fix software flaws.

Anthropic reviewed the technique and says the vulnerabilities it surfaced were minor and already publicly known. It claims other models, including OpenAI’s GPT-5.5, can discover the same flaws without any bypass at all, though TNW could not independently verify this specific comparison.

The company is drawing a sharp distinction between narrow jailbreaks, which can extract some information in specific circumstances, and universal jailbreaks, which broadly defeat a model’s safeguards. Anthropic says no tester has yet found a universal jailbreak for Fable 5, though the UK AI Safety Institute did develop a partial jailbreak for single-turn vulnerability queries within a few hours of testing.

Thousands of hours of red-teaming, then this

Before launch, Anthropic says it worked with the US government, the UK AISI, and multiple private third-party organisations to red-team Fable’s safeguards for thousands of hours. An external bug bounty reportedly produced no universal jailbreaks in over 1,000 hours of testing.

The company also adopted a controversial 30-day mandatory data retention policy for all Fable and Mythos traffic, overriding existing zero-retention agreements with customers. Anthropic says the retention is necessary to detect patterns of misuse that only become visible across multiple requests.

Anthropic fires back

In an unusually combative public statement, Anthropic said it disagrees that a narrow potential jailbreak should be “cause for recalling a commercial model deployed to hundreds of millions of people.” It warned that applying this standard across the industry would “essentially halt all new model deployments for all frontier model providers.”

The company called for any government power to block unsafe deployments to be grounded in “a statutory process that is transparent, fair, clear, and grounded in technical facts.” It said this action fails that test.

A pattern of tension

The directive lands against a backdrop of escalating friction between Anthropic and Washington. The Pentagon blacklisted Anthropic as a national security supply chain threat earlier this year, even as the NSA continued using Claude because no alternative existed.

Anthropic sued the government over that designation. The same administration that branded the company a security risk has also urged banks to adopt its technology and authorised the NSA to keep using Mythos on classified networks.

The contradiction highlights the government’s own uncertainty about how to handle a company whose AI capabilities it simultaneously fears and depends on.

What it means for the industry

If the precedent holds, any frontier AI model could face a similar recall over a single reported jailbreak, regardless of severity. That prospect is likely to send a chill through an industry that has spent the past year racing to deploy increasingly capable models with growing government scrutiny.

Anthropic says it believes the directive is a misunderstanding and is working to restore access as soon as possible. It has promised to share more details within 24 hours.