Anthropic’s Mythos found flaws in classified U.S. systems, official says

A U.S. official said Anthropic’s Mythos model identified vulnerabilities in highly sensitive, secure U.S. government computer systems during a testing exercise with U.S. intelligence agencies. The weaknesses were found within hours, though the official said that did not mean the model could exploit them in that time.

The work was part of Project Glasswing, an Anthropic initiative involving technology companies and other firms focused on protecting critical software from risks the Mythos model could pose to public safety, national security and the economy. Democratic Sen. Mark Warner referenced the testing during a Senate hearing, saying the capability had been described by Gen. Joshua Rudd, head of the National Security Agency and U.S. Cyber Command.

The cooperation comes amid rising tension between Anthropic and the Trump administration. The administration recently directed Anthropic to prevent foreign nationals from using Fable 5 and Mythos 5, while a new executive order created a voluntary process for reviewing national security risks from advanced AI systems before public release. More than 100 cybersecurity experts and executives urged the government to lift the directive, arguing that restricting defensive tools could benefit U.S. adversaries.