TL; DR
The US government issued an export control directive on June 12, ordering Anthropic to stop access to Fable 5 and Mythos 5, citing national security concerns due to the reported jailbreak. Anthropic complied, but publicly pushed back, calling the move disproportionate and warning it would stop all borderline model deployments if implemented industry-wide. The unprecedented recall comes just three days after Fable 5 was released.
The US government has ordered Anthropic to suspend access to two of its most capable AI models, the Fable 5 and Mythos 5, in what appears to be the first time Washington has forced a commercial AI product offline. The directiveThe lawsuit, which Anthropic said it received on June 12 at 5:21 p.m. ET, cites national security authorities as requiring the company to terminate the employment of all foreign nationals, including its own employees.
Because Anthropic cannot reliably differentiate foreign nationals from local users in real time, the practical result is a hard shutdown of both models for every customer worldwide. Access to all other models of Anthropic remains unaffected.
Three days from start to close
anthropic Fable 5 launched on June 9 as the first “Mythos-class” model to be released to the general public. Fable 5 is a public-facing version with security classifiers that redirect flagged requests to the weaker Claude Opus 4.8, while Mythos 5, the same basic model with cyber security raised, is limited to vetted cybersecurity defenders and critical infrastructure operators.
💜 of EU technology
The latest rumblings from the EU tech scene, a story from our wise founder Boris and some questionable AI art. Free in your inbox every week. Register now!
Both were models is already at the transition point of the Trump administrationhad previously acted to prevent Anthropic from spreading the wider Mythos. The company offered Fable 5 for free to Pro, Max, Team and Enterprise subscribers until June 22.
Jailbreak controversy
The Commerce Department reportedly took action after an unnamed company claimed to have found a way to jailbreak Mythos. Axios. Anthropic says it obtained verbal evidence of a narrow, non-universal jailbreak that essentially involves requiring the model to read a specific codebase and fix software flaws.
Anthropic reviewed the technique and said the vulnerabilities it found were minor and already publicly known. Claims including other models OpenAI’s GPT-5.5can detect the same defects without any bypass, although TNW cannot independently verify this particular comparison.
The company makes a sharp distinction between narrow jailbreaks that can extract some information in specific circumstances and universal jailbreaks that broadly violate model protections. Anthropic says no testers have yet found a universal jailbreak for the Fable 5, although the UK’s AI Security Institute develop a partial jailbreak for one-time vulnerability inquiries within hours of testing.
Thousands of hours of red-teaming, then this
Before launching, Anthropic said it worked with the U.S. government Great Britain AISIand many private third-party organizations to put together Fable’s keepers for thousands of hours. The external bug bounty has 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, canceling existing zero-retention contracts with customers. Anthropic says it should catch up detect patterns of abuse visible only among multiple requests.
Anthropogenic fires are making a comeback
In an unusually combative public statement, Anthropic said it disagreed that the narrow potential jailbreak was “reason to recall a commercial model deployed to hundreds of millions of people.” He warned that the industry’s implementation of this standard would “essentially halt all new model deployments for all frontier model providers”.
The company called on any government force to rely on a “transparent, fair, clear and technically fact-based legal process” to block unsafe postings. He said that this action does not pass that test.
A pattern of tension
The directive comes amid growing friction between Anthropic and Washington. Pentagon Blacklisted Anthropic as a national security supply chain threat Earlier this year, even the NSA continued to use Claude because there was no alternative.
anthropic sued the government on that destination. The same administration that called the company a security risk urged banks to adopt its technology and allowed the NSA to use Mythos on secret networks.
The contradiction underscores the government’s uncertainty about how to deal with a company that is simultaneously afraid of and dependent on AI’s capabilities.
What it means for the industry
If precedent holds, any borderline AI model could face a similar recall over a reported jailbreak, regardless of severity. The prospect is likely to chill an industry that has spent the past year racing to field increasingly capable models. increased government control.
Anthropic says it believes the directive was a misunderstanding and is working to restore access as soon as possible. He promised to share more detailed information within 24 hours.
