U.S. export controls force Anthropic to restrict access to AI models Mythos and Fable

The White House ordered Anthropic to restrict exports of its AI models Mythos and Fable to anyone outside the U.S. and foreign nationals inside the country, citing unspecified national security concerns. Anthropic complied within approximately 90 minutes, making both models unavailable to all users.

The directive followed two reported events: Amazon CEO Andy Jassy alerted the administration after Amazon researchers claimed to bypass safeguards in Fable 5, and Anthropic granted access to Mythos through its limited partner program to a South Korean telecom suspected of ties to China. Anthropic disputes the characterization of the bypass as a full jailbreak, calling it a narrow, already-patched issue.

The U.S. government has a decades-long history of attempting to control the spread of cybersecurity-related software through export controls, with mixed results. In the 1990s, the U.S. tried to block the distribution of encryption software like PGP, leading to the 'Crypto Wars' and eventual widespread adoption of end-to-end encryption.

During the early 2010s, governments expanded the Wassenaar Arrangement to include surveillance and hacking software as dual-use technologies requiring export licenses. However, the agreement has faced challenges, including non-adherence by countries like Israel and inconsistent enforcement by signatories such as Italy.

The impasse between Anthropic and the administration remains unresolved. Analysts suggest the administration may lift the restriction to maintain U.S. AI companies' global competitiveness, or AI labs may face new compliance burdens requiring government approval before serving foreign customers.