Anthropic suspends Claude Fable 5 and Mythos 5 after US directive

Anthropic says it has suspended access to Claude Fable 5 and Claude Mythos 5 after receiving a US government export-control directive on June 12, 2026. The company says the directive applies to access by any foreign national, whether inside or outside the United States, including foreign-national Anthropic employees. Anthropic’s practical conclusion was blunt: to comply, it says it must disable both models for all customers.

The suspension lands three days after Anthropic launched Fable 5 as a generally available, safeguarded Mythos-class model and Mythos 5 as a more restricted version for selected cyberdefenders and infrastructure partners. That turns a product launch into a policy test case. A frontier model did not merely get delayed before release. Anthropic says already-launched access has been pulled back because the government treated the model as an export-control object.

The access rule is the story

Anthropic says it received the directive at 5:21pm ET on June 12. The company says the letter did not provide specific details of the national security concern, but that Anthropic understands the government believes it became aware of a method for bypassing, or jailbreaking, Fable 5. Anthropic says it reviewed a demonstration involving a small number of previously known, minor vulnerabilities and argues that other publicly available models can discover similar issues without a bypass.

That is Anthropic’s account, not an official technical finding from the government. The difference matters. If a Commerce or BIS document becomes public, the facts may sharpen. For now, the source-of-record for the technical rebuttal is the company that is challenging the directive.

The operational effect is already clear. Anthropic says the order requires suspending access by foreign nationals, including foreign-national employees. In a globally staffed AI company and a globally sold AI product, that is not a simple geography toggle. It affects customers, internal staff, and cross-border support paths.

Fable and Mythos were designed as a split-access model

The June 9 launch post explains why Fable 5 and Mythos 5 were sensitive before the directive. Anthropic described Fable 5 as a Mythos-class model made safe for general use through safeguards. On some sensitive topics, Anthropic said requests would be routed to Claude Opus 4.8 instead. It also said those safeguards were intentionally conservative and would trigger in less than 5% of sessions on average.

Mythos 5 was the tighter-access version. Anthropic said it was the same underlying model as Fable 5, but with safeguards lifted in some areas for a small group of cyberdefenders and infrastructure providers. The company said Mythos 5 would initially be deployed through Project Glasswing in collaboration with the US government, with plans for a broader trusted-access program.

Pricing was also part of the launch. Anthropic listed both models at $10 per million input tokens and $50 per million output tokens, less than half the price of Claude Mythos Preview. Developers were told they could use claude-fable-5 through the Claude API.

That context makes the suspension more than a single-model availability note. Anthropic had tried to separate general access, safeguarded access, and trusted cyber access. The directive, as Anthropic describes it, collapsed that split into a full customer suspension.

The government action is broader than a takedown

AP described the move as the US government’s most significant step so far to restrict access to advanced AI models, and reported that the Commerce Department did not immediately respond to a request for comment. Axios reported that Commerce Secretary Howard Lutnick sent Anthropic CEO Dario Amodei a letter saying Mythos 5 and Fable 5 would be subject to export controls outside the United States and to all foreign persons within the country.

The key word is access. AI export controls have usually been discussed through chips, cloud capacity, model weights, or overseas deployment. This episode reaches the hosted model surface itself. If the rule path holds, the compliance burden is not just “who can buy the hardware?” It becomes “who can touch the model, from where, and under which employment or citizenship status?”

For enterprise buyers, that is the near-term lesson. A model that is available on Tuesday can become unavailable on Friday for reasons unrelated to price, latency, or model quality. The risk is not limited to Anthropic. Any team building around frontier models for security, scientific, or regulated work should assume model access may acquire policy constraints that are more granular than ordinary country availability.

What teams should check now

Teams using Claude should first confirm whether they were actually on Fable 5 or Mythos 5. Anthropic says other models are unaffected, so most ordinary Claude workflows may continue on Opus, Sonnet, or Haiku-class models. The risk sits with users who deliberately selected the newest Mythos-class capability or who were part of a trusted-access program.

The second check is contractual and geographic. If a workflow depends on a model that could be export-controlled, the team should know which employees, vendors, and support staff can access prompts, logs, outputs, evaluation harnesses, and admin consoles. “US customer” and “US-only access” are not the same thing when foreign-national access inside the United States is part of the stated restriction.

The third check is substitution. Anthropic says access may be restored and that it is working to resolve what it calls a misunderstanding. That may happen quickly. But production owners should still document which fallback model is acceptable, which tests prove the fallback is good enough, and which sensitive workflows should pause instead of silently degrading.

The next 24 hours matter

Anthropic said it would share more details over the next 24 hours. The most important missing pieces are the legal basis, the exact license requirements, whether US-only access can be restored, and whether the government publishes a technical explanation for the jailbreak concern. Without those, this remains a one-sided public record: Anthropic’s statement, plus reporting that describes the government’s action.

Still, the shape of the precedent is already visible. Frontier model deployment is moving from model-card risk management into export-control and national-security process. The AI Feed’s read is that the operational takeaway is lifecycle discipline: teams need model-access inventories, fallback plans, and a policy-risk owner for frontier systems that matter to production work.

For readers tracking Anthropic’s model strategy, see our AI model leaderboard and AI company tracker.