Anthropic announced that it has suspended access to its advanced AI models, Fable 5 and Mythos 5, after receiving an export control directive from the U.S. government citing national security authorities. The directive requires the company to immediately halt access to the models for all foreign nationals, both within and outside the United States, including foreign-national employees of Anthropic.

The order, received at 5:21 p.m. ET on June 12, leaves Anthropic with no alternative but to disable access to both models globally to ensure full compliance. Access to all other Anthropic AI models remains unaffected.

“The government’s directive requires us to suspend access to Fable 5 and Mythos 5 immediately,” Anthropic said in a statement. “While we are complying with this legal order, we disagree with the basis for the action and believe it reflects a misunderstanding of the technical evidence.”

According to Anthropic, government officials indicated concerns regarding a potential method of bypassing, or “jailbreaking,” the safeguards built into Fable 5. The company stated that it reviewed demonstrations of the reported technique and found that it revealed only a limited number of previously known, relatively minor vulnerabilities. Anthropic further noted that comparable findings can be generated using other publicly available AI models without requiring any safeguard bypass.

Extensive Safety Testing and Safeguards

Anthropic emphasized that Fable 5 underwent extensive safety evaluations before launch, including thousands of hours of red-team testing conducted in collaboration with U.S. government agencies, the UK AI Safety Institute, independent third-party organizations, and internal security teams.

The company stated that these assessments demonstrated safeguards that were substantially stronger than those of previously deployed AI systems and that no evaluator had identified a universal jailbreak capable of broadly circumventing the model’s cybersecurity protections.

Recognizing that perfect jailbreak resistance may not be achievable for any AI model, Anthropic said it adopted a defense-in-depth security strategy. This approach combines layered safeguards, continuous monitoring, rapid response mechanisms, and a 30-day customer data retention policy designed to facilitate the detection and mitigation of emerging threats.

“Every safeguard currently used across the AI industry is vulnerable to some form of narrow, non-universal jailbreak,” the company noted. “Our objective has been to make such exploits either highly limited in scope or prohibitively difficult to develop, while maintaining robust monitoring capabilities.”

Company Challenges Technical Basis of Directive

Anthropic stated that it has not been provided with evidence of any harmful outcomes resulting from the reported vulnerabilities. The company characterized the disclosed examples as benign or limited findings that do not provide meaningful capabilities beyond those already available through existing AI systems.

The company further asserted that the reported capabilities cited by government officials are broadly available from multiple AI models currently deployed across the industry and are routinely used by cybersecurity professionals to identify and remediate software vulnerabilities.

Anthropic indicated that it intends to provide additional technical details regarding the findings within the next 24 hours.

Broader Implications for AI Innovation

While reaffirming its support for government oversight of advanced AI systems, Anthropic expressed concern that the directive could establish a precedent that would significantly hinder innovation across the AI sector.

“We believe governments should have the authority to prevent genuinely unsafe deployments through transparent, fair, and technically grounded processes,” the company said. “However, recalling a commercially deployed model based on a narrow and unverified potential jailbreak does not meet those standards.”

Anthropic warned that applying similar criteria across the industry could effectively halt the deployment of future frontier AI systems.

Commitment to Customers

The company apologized to customers affected by the sudden suspension and stated that it is actively engaging with government authorities in an effort to resolve the issue and restore access as quickly as possible.

“We recognize the disruption this causes for our users and partners,” Anthropic said. “We remain committed to maintaining the highest standards of safety and security while ensuring continued innovation in advanced AI systems.”