US government orders Anthropic to suspend access to Fable 5 and Mythos 5
Key Points
- Anthropic disabled its Fable 5 and Mythos 5 AI models for all customers on 12 June 2026 to comply with a US government export control directive citing national security.
- The directive ordered access suspended for any foreign national; Anthropic said it disabled the models for everyone because it cannot separate foreign nationals from other users in real time.
- All of Anthropic's other AI models remain available and unaffected.
- The order is understood to relate to a potential method of bypassing, or jailbreaking, Fable 5, though Anthropic says the vulnerabilities involved are minor and discoverable by other models.
- Anthropic says it is complying but disputes the basis for the order, calling it a misunderstanding and saying it is working to restore access.
Anthropic has disabled access to its two most powerful AI models, Fable 5 and Mythos 5, for all customers after receiving an export control directive from the US government.
The company said it received the directive on Friday (12 June) at 5:21pm Eastern Time.
According to Anthropic, the order cited national security authorities and required the firm to suspend all access to the two models for any foreign national, whether inside or outside the United States, including the company’s own foreign national employees.
Anthropic said the directive did not set out the specific national security concern behind it.
Why are all customers affected?
Although the order targets foreign nationals, Anthropic said it could not reliably separate those users from the rest of its customer base in real time.
As a result, the company said it had to disable Fable 5 and Mythos 5 for every customer worldwide to ensure compliance. Anthropic said access to all of its other models remains unaffected.
The two models were released earlier in the week, on Tuesday (9 June), and were described by Anthropic at launch as the most powerful systems it had shared.
Both were built on the same technical foundation. Only Fable 5 was made available to the general public, with stronger restrictions on the questions users can ask, particularly on cybersecurity and biology.
Mythos 5 was released without those safeguards to a select group of trusted partners, including cybersecurity and infrastructure companies.
The jailbreak concern
Anthropic said its understanding was that the government believed it had identified a method of bypassing, or “jailbreaking”, Fable 5.
The company said it reviewed a demonstration of the technique, which it said identified a small number of previously known, minor vulnerabilities.
According to Anthropic, those vulnerabilities were relatively simple and could be discovered by other publicly available models without any bypass.
Anthropic said the potential jailbreak shared with the government essentially involved asking the model to read a specific codebase and fix software flaws.
The company said it had validated that the level of capability shown was widely available from other models, including OpenAI’s GPT-5.5, and was used daily by security defenders.
Anthropic said it was complying with the legal directive but disagreed that a narrow potential jailbreak should justify recalling a commercial model used by hundreds of millions of people.
The company said that if such a standard were applied across the industry, it would effectively halt new model deployments by all frontier model providers.
The firm reiterated its view that the government should be able to block unsafe deployments through a statutory process that is transparent, fair, clear, and grounded in technical facts, and said this action did not meet those principles.
Anthropic said it believed the directive stemmed from a misunderstanding and that it was working to restore access as soon as possible.
The models were built on Claude Mythos Preview, an earlier system used for security research through a programme called Project Glasswing. Anthropic has previously said participants used it to identify and fix security issues, with Mozilla reporting that it resolved hundreds of vulnerabilities with the model’s help.
