Anthropic: U.S. Government Orders Global Suspension of Fable 5 and Mythos 5, Citing National Security

The U.S. government, on June 12, 2026, at 5:21 PM Eastern Time, delivered to Anthropic an export control directive ordering the immediate global suspension of access to Claude Fable 5 and Claude Mythos 5. The directive arrived just three days after both models became commercially available — they launched on June 9, 2026. The measure covers all users without exception, including Anthropic employees who are not U.S. citizens. Access to all other models in Anthropic’s portfolio remains unchanged.

Content of the Government Directive

The United States government invokes its authority under national security legislation and asserts that a method of bypassing the safety mechanisms of the Fable 5 model has been identified. According to the government, this method could enable misuse of the model in ways that pose a security risk. The directive specifies no timeline or conditions under which the suspension could be lifted, nor does it outline any review process.

The scope of the measure is without precedent in the commercial AI industry: the ban is immediate and global, leaving no room for a transitional period or for protective mechanisms that might mitigate the business impact on clients who, within just a few days of the launch, had already begun integrating Fable 5 and Mythos 5 into their own processes or had signed business contracts for them.

Is a Single Jailbreak Sufficient Grounds for a Global Suspension?

The core dispute between Anthropic and the government concerns the severity and scope of the demonstrated security flaw. According to Anthropic’s statement, the government demonstrated “a narrow, non-universal jailbreak that essentially amounts to asking the model to read a specific codebase and correct software vulnerabilities.” The company notes that this capability is “widely available in other models, including OpenAI’s GPT-5.5.”

Anthropic’s experts confirm that the demonstration revealed “a small number of previously known, minor vulnerabilities” — flaws that are, however, present in other publicly available models as well. The key distinction in Anthropic’s argument is the difference between a non-universal and a universal jailbreak: the company emphasizes that no universal jailbreak, one that would function consistently across all users and in all contexts, was found even in thousands of hours of testing.

The Fable 5 and Mythos 5 models were tested with the participation of the U.S. government, the UK’s AISI (AI Safety Institute), and a range of independent third parties. Anthropic describes the resulting safety mechanisms as “significantly more effective than those of any previously deployed model” and notes that, as of the date of the directive, no harmful consequences from known jailbreaks had been recorded.

Anthropic’s Public Response

Anthropic is complying with the directive but publicly and unambiguously contests its justification. In a written statement, the company argues that the discovery of a single narrow jailbreak does not justify a global suspension of models used, in Anthropic’s words, by “hundreds of millions of users.” The company calls for the establishment of “a transparent, fair, and statutory review process grounded in technical facts.”

Particularly problematic for Anthropic is the complete absence of prior consultation. The company notes that it was not informed or consulted within a reasonable timeframe that would have allowed it to propose alternative security solutions or technical risk-mitigation measures. This lack of transparency and procedural fairness, Anthropic argues, serves no one’s interests — neither those of safety advocates nor of the industry itself.

Anthropic also warns of potential systemic consequences: applying the same standard consistently across the entire AI industry, the company states, “would essentially halt all new model deployments.” Such an outcome, Anthropic contends, would in the long run harm both national technological capacity and the very security goals cited in the directive.

Consequences for Users and the Broader AI Industry

For end users and business clients who had begun using or integrating Fable 5 and Mythos 5 within just three days of the launch, the suspension arrives without prior notice and with no clear resolution timeline. Anthropic’s statement provides no specific estimates of the number of affected users, nor any communication plans for business partners and development teams that had been building on these models.

At a broader level, this situation raises a series of questions that regulators, legislators, and the industry will need to resolve together: what threshold of security flaw justifies such measures, who has the right to define that threshold, and through what transparent process such decisions should be made. Anthropic publicly calls for the establishment of clear, technically grounded standards for future cases of this kind — not only in the United States, but at the level of international regulation of AI systems.