Anthropic’s Mythos model successfully identified vulnerabilities in sensitive US government computer systems during a collaborative testing exercise, according to an official cited by the Associated Press. This testing was conducted under Project Glasswing, an initiative designed to secure critical infrastructure against potential threats to national security, public safety, and the economy. US Senator Mark Warner stated during a June 11 Senate committee hearing that the AI model breached classified systems within hours rather than weeks, a finding he attributed to information from General Joshua Rudd, head of the National Security Agency and US Cyber Command. Neither the NSA nor Anthropic provided official comment on these specific findings.

Growing tensions persist between Anthropic and the Trump administration despite this recent cooperation. The administration issued a directive earlier in June requiring Anthropic to restrict foreign nationals from accessing its latest models, specifically Fable 5 and Mythos 5. This mandate followed an executive order signed by President Donald Trump that establishes a voluntary framework for the federal government to review the national security risks of advanced AI systems for up to a month before public release.

Anthropic has disabled these models for all customers to comply with the directive, though the company stated it does not believe the government's measures are justified by identified security concerns. A group of over 100 cybersecurity executives and experts, including leaders from Adobe and Nvidia, have formally requested that the administration lift these restrictions. The group argues in a letter that while Mythos models are effective at detecting and weaponizing software flaws, they are not uniquely skilled in these areas, and removing advanced defensive tools could potentially assist foreign adversaries while leaving US systems vulnerable.

Anthropic’s Mythos model successfully identified vulnerabilities in sensitive US government computer systems during a collaborative testing exercise, according to an official cited by the Associated Press. This testing was conducted under Project Glasswing, an initiative designed to secure critical infrastructure against potential threats to national security, public safety, and the economy. US Senator Mark Warner stated during a June 11 Senate committee hearing that the AI model breached classified systems within hours rather than weeks, a finding he attributed to information from General Joshua Rudd, head of the National Security Agency and US Cyber Command. Neither the NSA nor Anthropic provided official comment on these specific findings.

Growing tensions persist between Anthropic and the Trump administration despite this recent cooperation. The administration issued a directive earlier in June requiring Anthropic to restrict foreign nationals from accessing its latest models, specifically Fable 5 and Mythos 5. This mandate followed an executive order signed by President Donald Trump that establishes a voluntary framework for the federal government to review the national security risks of advanced AI systems for up to a month before public release.

Anthropic has disabled these models for all customers to comply with the directive, though the company stated it does not believe the government's measures are justified by identified security concerns. A group of over 100 cybersecurity executives and experts, including leaders from Adobe and Nvidia, have formally requested that the administration lift these restrictions. The group argues in a letter that while Mythos models are effective at detecting and weaponizing software flaws, they are not uniquely skilled in these areas, and removing advanced defensive tools could potentially assist foreign adversaries while leaving US systems vulnerable.