Anthropic Cuts Off Access to Fable 5 and Mythos 5 Over US Directive

Anthropic said it has suspended access to its Fable 5 and Mythos 5 AI models after receiving a U.S. government export control directive tied to national security concerns. The company disabled access for all users, including foreign nationals, effective immediately after it received the order at 5:21 p.m. ET on Friday.

In a statement posted on its website, Anthropic said the directive instructed it to suspend “all access” to Fable 5 and Mythos 5 by any foreign national, regardless of whether they are located inside or outside the United States. The company also said its other models, such as Opus 4.8, are not affected by the order.

Key takeaways

• Anthropic suspended global access to Fable 5 and Mythos 5 after a U.S. export control directive citing national security concerns.
• The order was received at 5:21 p.m. ET and required the company to halt access “by any foreign national,” including foreign national employees.
• Other Anthropic models, including Opus 4.8, are reported as unaffected.
• Anthropic says the government provided only verbal evidence of a narrow, non-universal jailbreak risk.
• The company disputes the premise that such a finding should trigger a recall-style response for models used at large scale.

Export-control order forces worldwide suspension

According to Anthropic, the directive it received compelled the company to remove access to Fable 5 and Mythos 5 for all users. Anthropic said it acted abruptly to comply with the government’s legal instruction and ensure the directive’s requirements are met.

While the company did not provide additional details in the statement about the specific nature of the alleged threat, it said it understands authorities are concerned about a potential “jailbreak” method that could bypass the models’ safeguards.

What Anthropic says the government’s concern is

Anthropic characterized the government’s evidence as limited and not universal in scope. The company said the government did not provide detailed technical information but instead offered “verbal evidence” of a potential narrow, non-universal jailbreak. Anthropic described that type of jailbreak as involving requests for the model to read a specific codebase and then fix software flaws found there.

In Anthropic’s framing, a non-universal jailbreak is fundamentally different from a “universal jailbreak”—the latter would enable broad, repeatable bypasses that work across many contexts. Anthropic argued that the alleged risk, as presented to the company, appears narrower than what would be required for a universal safeguard failure.

The company also expressed disagreement with treating a narrow, potential jailbreak as justification to roll back a frontier model that it said is deployed at very large scale. Anthropic argued that if the same threshold were applied industry-wide, it would effectively stop new frontier model deployments.

Timeline: new releases followed by compliance shutdown

The suspension came only days after Anthropic released Fable 5 and Mythos 5. The release followed the company’s earlier work with “Mythos Preview,” a general-purpose language model that Anthropic has previously said identified thousands of vulnerabilities in critical software.

Earlier coverage noted that Fable 5 and Mythos 5 were built on top of Mythos Preview and designed to demonstrate advanced capabilities, including security-relevant behaviors. The company’s current statement indicates that the export-control directive arrived in the middle of this rollout cycle, leaving Anthropic to disable access immediately rather than adjust the models incrementally.

Anthropic also said it believes the government’s order may stem from a misunderstanding and that it is working toward restoring access for users as soon as possible.

Why this matters beyond AI product access

For investors, developers, and users, the episode highlights how quickly geopolitics and compliance requirements can intersect with frontier AI deployment. Even when a company believes an identified risk is narrow and non-universal, a government directive can still force immediate operational changes—particularly when export-control rules are interpreted to cover access by foreign nationals.

The situation also underscores a practical tension in model governance: technical risk assessments can point to targeted mitigation, while legal directives may impose broad suspension measures to ensure compliance. Anthropic’s claim that other models are not affected suggests the company may be trying to isolate the impacted releases, but the pathway back to normal availability is uncertain and depends on the government’s engagement.

Going forward, market participants and AI users will likely watch for any update on what additional evidence or clarification the government provides, whether Anthropic can demonstrate that the risk is contained, and how quickly access can be restored to Fable 5 and Mythos 5 without repeating the compliance-triggering conditions.