Researchers at Amazon reportedly conducted a jailbreak on Anthropic's artificial intelligence models, leading to export restrictions imposed by the U.S. government. According to the Wall Street Journal, the Commerce Department restricted access to Anthropic's Fable 5 and Mythos 5 models for foreign nationals and users outside the United States. Following the jailbreak, Amazon CEO Andy Jassy and White House officials, including Treasury Secretary Scott Bessent, met to discuss the situation, resulting in a decision to limit access to the AI tools.
Katie Moussouris, CEO of cybersecurity firm Luta Security, stated that the jailbreak research involved prompting Anthropic’s model to reveal information about security vulnerabilities. She emphasized that the information would be more beneficial for cybersecurity professionals than for malicious actors. Moussouris questioned the White House's assessment of the situation, calling it an overreaction.
Amazon did not confirm its involvement in the jailbreak testing but noted that it regularly advises the government on cybersecurity matters. An Amazon spokesperson stated that it is common for governments to seek their counsel on potential security risks.
Commerce Secretary Howard Lutnick informed Anthropic CEO Dario Amodei that the models would be subject to export controls. Anthropic believes the government's decision was linked to the jailbreak technique, which it described as revealing a small number of known vulnerabilities. The company stated that other publicly available models could also identify these vulnerabilities without requiring a bypass.
Anthropic is complying with the government's directive to remove access to the models but disagrees with the rationale behind the order, asserting that the findings should not warrant recalling a widely deployed commercial model.