The US has instructed Anthropic to discontinue Fable 5 and Mythos 5.

      TL;DR: On June 12, the US government mandated that Anthropic suspend access to its advanced AI models, Fable 5 and Mythos 5, due to national security issues stemming from an alleged jailbreak. Anthropic complied but publicly criticized the decision as excessive, warning that such actions could lead to a halt in all frontier model rollouts if they became widespread. This unusual recall followed just three days after the launch of Fable 5.

      The US government has instructed Anthropic to discontinue access to its two advanced AI models, Fable 5 and Mythos 5, marking a historic occurrence where Washington has mandated the removal of a commercial AI product. The directive, received by Anthropic at 5:21 PM ET on June 12, references national security concerns and requires the company to block access for all foreign nationals, including its own staff.

      Since Anthropic cannot effectively distinguish between foreign and domestic users in real time, this means both models are inaccessible to all customers globally. Nonetheless, access to Anthropic’s other models remains unaffected.

      From launch to shutdown in three days

      Anthropic released Fable 5 on June 9 as its inaugural “Mythos-class” model available to the public. Fable 5 is designed for public use, featuring safety classifiers that redirect flagged requests to the less powerful Claude Opus 4.8, while Mythos 5, which has enhanced capabilities with security measures removed, was limited to authorized cybersecurity professionals and critical infrastructure operators.

      The EU tech scene's latest developments, insights from our wise founder Boris, and some questionable AI artwork are available every week through our free newsletter. Sign up now! Both models were already under scrutiny from the Trump administration, which had previously attempted to restrict Anthropic’s broader Mythos launch. The company had been offering Fable 5 free to Pro, Max, Team, and Enterprise subscribers until June 22.

      The jailbreak controversy

      The Commerce Department reportedly took action after an unnamed company claimed to have discovered a way to jailbreak Mythos, as reported by Axios. Anthropic stated it only received verbal confirmation of a narrow, specific jailbreak that involves asking the model to analyze a particular codebase and rectify software errors.

      After examining the technique, Anthropic concluded that the vulnerabilities identified were minor and already publicly known. It contends that other models, including OpenAI’s GPT-5.5, can uncover similar flaws without the need for any bypass. However, TNW could not independently verify this specific claim.

      The company is emphasizing a clear distinction between narrow jailbreaks, which may extract some information under specific conditions, and universal jailbreaks, which broadly undermine a model’s safeguards. Anthropic asserts that no tester has yet found a universal jailbreak for Fable 5; however, the UK AI Safety Institute did develop a partial jailbreak for specific vulnerability queries shortly after testing began.

      Extensive red-teaming efforts

      Prior to launch, Anthropic claims to have collaborated with the US government, the UK AISI, and various private third-party organizations to scrutinize Fable's safeguards for thousands of hours. An external bug bounty reportedly yielded no universal jailbreaks in over 1,000 hours of testing.

      Additionally, the company implemented a controversial 30-day mandatory data retention policy for all Fable and Mythos traffic, overriding previous zero-retention agreements with clients. Anthropic argues that this retention is essential to identify patterns of misuse that become apparent only through numerous requests.

      Anthropic responds

      In a notably defiant public statement, Anthropic asserted that a narrow potential jailbreak should not warrant the recall of a commercial model accessed by hundreds of millions of users. The company cautioned that applying this standard industry-wide would "effectively halt all new model deployments for all frontier model providers."

      Anthropic called for government powers to block unsafe deployments to be based on "a statutory process that is transparent, fair, clear, and grounded in technical facts," claiming that this action does not meet that criterion.

      An ongoing pattern of tension

      This directive emerges amidst rising tensions between Anthropic and the US government. Earlier this year, the Pentagon labeled Anthropic a national security supply chain risk, even as the NSA continued to utilize Claude due to the lack of alternatives. Anthropic subsequently filed a lawsuit against the government over that designation. The same administration that deemed the company a security threat has encouraged banks to adopt its technology and authorized the NSA to continue using Mythos on secure networks.

      This contradiction highlights the government’s own uncertainty in managing a company whose AI capabilities it simultaneously fears and relies upon.

      Implications for the industry

      If this precedent is established, any frontier AI model could potentially face a similar recall over a single reported jailbreak, irrespective of its severity. This possibility is likely to alarm an industry that has spent the past year racing to deploy increasingly advanced models under heightened government scrutiny.

      Anthropic believes the directive stems from a misunderstanding and is working to restore access as quickly as possible. The company