U.S. Bans Anthropic from Exporting Mythos and Fable AI Models

The White House last Friday ordered Anthropic to restrict exports of its Mythos and Fable AI models to anyone outside the United States and foreign nationals inside the country. Both models have been offline for one week following the Commerce Department directive.

What Triggered the Ban

Two events led to the export control order, according to reports:

1. Anthropic granted access to Mythos through its limited partner program to a South Korean telecom, widely reported to be SK Telecom. U.S. officials suspected the company had ties to China, which SK Telecom has denied.

2. Amazon CEO Andy Jassy alerted the administration after Amazon researchers claimed to find a way around Fable 5's safeguards. Anthropic disputes the "jailbreak" label, calling it a narrow, already-patched issue rather than a complete defeat of safety measures.

Anthropic had 90 minutes to comply with the directive and immediately restricted access to both models.

About Mythos

Anthropic launched Mythos in April and has marketed it as a cybersecurity tool with potential for significant harm if widely distributed. Before the ban, only approximately 150 vetted companies and government organizations had access. The stated goal was helping defenders secure software before malicious actors could develop similar capabilities.

Context window, pricing, and benchmark scores for Mythos and Fable have not been publicly disclosed.

Historical Precedent for Cyber Export Controls

The U.S. government has attempted to control cyber technology exports for decades with mixed results:

1990s Encryption Controls: The U.S. Customs Service investigated PGP creator Phil Zimmermann for allegedly violating arms export controls. Zimmermann published PGP's source code as a printed book, sparking the "Crypto Wars." The investigation was eventually closed, enabling widespread adoption of encryption technologies now used by billions.

2010s Spyware Controls: Governments expanded the Wassenaar Arrangement to classify surveillance software as dual-use technology requiring export licenses. However, major spyware-producing countries like Israel don't adhere to the agreement. European enforcement has been inconsistent—Italy granted export licenses to Hacking Team despite sales to oppressive governments. Spyware makers like Intellexa have relocated to countries with lax controls.

Germany-based FinFisher shut down in 2022 after a multi-year investigation for allegedly selling spyware to Turkey without an export license.

Current Status

The standoff between Anthropic and the Trump administration continues. Two possible outcomes:

1. The administration lifts restrictions to keep American AI companies competitive globally—acknowledging that AI labs elsewhere, including China, will likely reach similar capabilities regardless of U.S. controls.

2. American AI companies require government approval before serving foreign customers, creating a compliance burden that would impact revenue.

What This Means

This represents the first major test of AI export controls and could establish precedent for how the U.S. regulates frontier AI systems. The historical failures of export controls for encryption and spyware suggest that restricting software distribution is difficult to enforce and often counterproductive. If the ban remains, it may push AI development to jurisdictions with fewer restrictions while hampering U.S. companies' global competitiveness. The outcome will likely shape compliance requirements for all major AI labs operating in the United States.