Anthropic’s highly advanced Mythos AI model has become the center of a growing debate over artificial intelligence and national security after the U.S. government ordered the company to suspend access to its Fable 5 and Mythos 5 models.
The Trump administration issued the directive under national security authorities, requiring Anthropic to halt access to the models for foreign nationals both inside and outside the United States. The restrictions also apply to foreign national employees working at Anthropic. To ensure compliance, the company disabled access to the models for all customers while leaving the availability of its other AI systems unchanged.
The action represents one of the most aggressive interventions by the U.S. government regarding frontier AI models and highlights growing concerns about the cybersecurity capabilities of advanced artificial intelligence systems.
Anthropic said it received the directive at 5:21 p.m. ET on June 12 and that the government did not provide specific details regarding its national security concerns. Based on information provided verbally, the company said officials believe they had identified a method of bypassing, or “jailbreaking,” the safeguards protecting Fable 5.
The company reviewed the reported technique and said it identified only a small number of previously known vulnerabilities, which Anthropic characterized as minor and relatively simple. According to the company, other publicly available models, including OpenAI’s GPT-5.5, can identify the same vulnerabilities without requiring any bypass methods.
Anthropic said that before launching Fable 5, it spent thousands of hours working with the U.S. government, the United Kingdom’s AI Security Institute, private organizations, and internal teams to conduct extensive red-team testing. According to the company, those tests demonstrated that Fable’s safeguards were stronger than those of previously deployed AI models.
The company noted that no testers had discovered a universal jailbreak capable of broadly bypassing the model’s protections across a wide range of cyber capabilities. Anthropic acknowledged that perfect resistance to jailbreaks may not currently be possible for any AI developer and said it adopted a “defense in depth” strategy designed to make attacks narrow in scope or prohibitively expensive.
As part of that approach, Anthropic implemented a 30-day customer data retention policy for Fable 5, enabling researchers to detect and mitigate successful attacks. The company said the policy carried high commercial costs but was necessary to improve security.
Anthropic said the only evidence it has received from the government concerns a narrow, non-universal jailbreak involving requests for the model to review a codebase and identify software flaws. The company stated that the resulting capabilities are widely available across the industry and are routinely used by cybersecurity professionals to protect systems.
The company added that it supports government oversight of unsafe AI deployments but believes any such process should be transparent, technically grounded, and based on clearly defined standards.
Anthropic apologized for the disruption to customers and said it believes the situation stems from a misunderstanding and that it is working to restore access as quickly as possible.
The controversy underscores the growing tension between rapid advances in AI and governments’ increasing efforts to regulate technologies that are becoming strategically important. As frontier models gain greater capabilities in areas such as software engineering and cybersecurity, policymakers are grappling with how to balance innovation with national security risks.
KEY QUOTES:
“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.”
“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
