Anthropic suspends Fable 5 & Mythos 5 after US order

Anthropic has suspended access to its Fable 5 and Mythos 5 models after receiving a US government export control directive. The order applies to all foreign nationals, including Anthropic employees who are not US citizens.

To comply, Anthropic disabled both models for all customers. Access to its other models remains unchanged.

According to the company, the US government cited national security authorities in its order but did not provide specific details about the concern behind the decision.

Anthropic said it understood that officials believe a method exists to bypass, or jailbreak, Fable 5. After reviewing a demonstration of the technique, it concluded that it exposed a small number of previously known vulnerabilities.

Those vulnerabilities appeared relatively simple, Anthropic said, adding that other publicly available models could identify them without any bypass method.

Government order

The directive suspends access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the US. Anthropic said the order also covers its own foreign national employees.

In describing the issue, Anthropic distinguished between what it called narrow, non-universal jailbreaks and a broader method that could override safeguards across a wide range of uses. It said testers had not found a universal jailbreak for Fable 5.

Before launch, Anthropic said it worked with the US government, the UK AI Safety Institute, private third-party organisations and internal teams to test the model's safeguards. Those exercises ran for thousands of hours, it said.

Anthropic argued that no model provider can currently make models perfectly resistant to jailbreaks. It said every safeguard used across the industry remains vulnerable in some circumstances to narrower forms of bypass.

Disputed findings

Anthropic said the government had given it verbal evidence of what it described as a potential narrow, non-universal jailbreak. According to the company, the example involved asking the model to read a specific codebase and fix software flaws.

It also said it had reviewed a report that it believes formed the basis of the directive. Anthropic argued that the level of capability described in that report was widely available from other models, including OpenAI's GPT-5.5, and routinely used by cyber defenders.

Anthropic said it had not received disclosure of a concerning non-universal jailbreak that led to a harmful result. It added that the potential jailbreaks disclosed to it were either benign responses or minor findings that offered no Mythos-specific uplift.

Safeguards debate

Before the suspension, Anthropic had presented Fable 5 as a model with strong safeguards designed to limit misuse in areas including cybersecurity. It said some users had complained that those protections were too broad.

Anthropic said it adopted a defence-in-depth approach for Fable 5, aiming to make jailbreaks either narrow or expensive to produce, while using monitoring to detect and shut down successful attacks. It added that this approach was also why it required 30-day retention of customer data for Fable 5.

That policy carried commercial costs, Anthropic said, but allowed it to research and mitigate jailbreaks. It maintained that the strategy reduced the risks posed by Fable 5 to a level comparable with existing models already deployed across the sector.

The suspension is likely to sharpen debate over how governments should intervene in frontier AI releases, especially when concerns focus on model safeguards rather than alleged direct harm. Anthropic said it supports a statutory process that is transparent, fair, clear and grounded in technical facts when the government seeks to block unsafe deployments.

"We are complying with the government's legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers. As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear and grounded in technical facts. This action does not adhere to those principles. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible," Anthropic said.