The Commerce Department invoked national security export controls late Friday to restrict the use of Anthropic’s AI models, Fable 5 and Mythos 5. The decision applies to foreign nationals both inside and outside the United States, including Anthropic’s own non-citizen employees. As a result, the company disabled the models for all users, claiming it had no alternative.
Defense-Oriented Research Sparks Controversy
Anthropic stated that the export controls were triggered by research into a potential "jailbreak" of its AI systems, which sought to bypass safeguards. However, Katie Moussouris, CEO of cybersecurity firm Luta Security, disputed this characterization. Speaking to the Wall Street Journal, Moussouris explained that the research conducted by Amazon focused on Defense Oriented Prompting (DOP), a method aimed at identifying security vulnerabilities for defensive purposes. "It’s not a jailbreak. It was DOP, capabilities defenders need," she asserted. Moussouris criticized the Commerce Department’s decision, stating, "If national defense is the goal, this just scored an own goal against us."
Implications for AI Development
Ben Murphy, a scholar at the Institute for Progress, raised concerns about the broader implications of the administration’s directive. He described the move as "another step on the balkanization of technology" and warned that it could lead AI developers to withhold models from public release or disengage from government collaboration. "Labs might be less inclined toward engaging with the government about potential vulnerabilities," Murphy noted.
"This action does not adhere to principles of transparency and fairness." — Anthropic Statement
Anthropic argued that while the government has the authority to restrict unsafe AI, such decisions should follow a "statutory process that’s transparent, fair, and based on technical facts." The company’s relationship with the administration has been strained, with the government previously labeling Anthropic a supply-chain risk for Pentagon contractors. Despite these tensions, Anthropic provided early access to the Mythos model, underscoring its commitment to addressing potential cybersecurity concerns.