U.S. Government Orders Anthropic to Disable Fable 5 and Mythos 5 Over National Security Concerns

The U.S. Commerce Department on Friday ordered Anthropic to immediately suspend all foreign access to Fable 5 and Mythos 5, its two most advanced AI models, citing national security concerns tied to a reported method of bypassing safety guardrails. The directive forced Anthropic to disable both models worldwide that evening, just three days after their launch.

Anthropic received the order at 5:21 p.m. ET via a letter from Secretary of Commerce Howard Lutnick to CEO Dario Amodei. The models had been released Tuesday with Anthropic calling them the most capable systems it had ever deployed. Mythos 5 was offered through Project Glasswing, a program that gave selected cybersecurity companies access to identify security flaws. It remains unclear how the order affects that initiative.

Government cited a narrow jailbreak; Anthropic called it widely available

The Commerce Department did not detail the specific national security concern. In a blog post Friday, Anthropic said it understands the government became aware of a technique for "jailbreaking" Fable 5, meaning methods that circumvent built-in safety controls. According to Anthropic, the government provided only verbal evidence of a "narrow, non-universal jailbreak" that involved prompting the model to read a specific codebase and find software flaws.

Anthropic disputed the severity. The company said it reviewed a report it believes formed the basis of the directive and found that the capabilities were already available in other publicly accessible models, including OpenAI's GPT-5.5. It noted those same capabilities are routinely used by cybersecurity professionals for defensive purposes.

• Anthropic acknowledged that perfect jailbreak resistance is impossible. It said Fable 5 used a "defense in depth" strategy combining narrow jailbreak resistance with active monitoring.
• No testers found a universal jailbreak capable of broadly bypassing the model's safeguards, according to Anthropic.
• The company argued the standard would halt new model deployments industry-wide if applied consistently.
• Katie Moussouris, CEO of Luta Security, attributed the issue to "Defense Oriented Prompting," a security-first method that treats natural language as code.
• Some reports suggested Amazon flagged the security issues, but Amazon did not respond to requests for comment.

Wider context: Anthropic and the Trump administration have a history of friction

The directive follows a prolonged dispute between Anthropic and the administration. In February, President Trump barred Anthropic's products from federal agencies after the company sought stronger restrictions on Pentagon use of its technology. Despite that, Anthropic gave the National Security Agency access to Mythos 5 for offensive cyber operations. Earlier this month, Trump signed an executive order directing agencies to bolster cyber defenses and establish a voluntary mechanism for early government access to powerful models before deployment.

Industry reaction was critical. Dean Ball of the Foundation for American Innovation called the move "baffling." Chris McGuire of the Council on Foreign Relations called the broad restriction and deemed export provisions "just absurd." Aaron Levie, CEO of Box, described it as "a big turning point for AI regulation," warning it sets a precedent where the government can deem specific models too powerful for certain uses. Anthropic executives scrambled to undo the action but have publicly accepted compliance. The broader implications for frontier model deployment remain uncertain as other providers watch closely.