WASHINGTON — Anthropic's Mythos artificial intelligence model identified vulnerabilities in classified U.S. government computer systems during a testing exercise with U.S. intelligence agencies in June 2025. Following the event, the U.S. government issued export controls and revoked federal access to the model.

A U.S. official stated that Anthropic teamed up with intelligence agencies to conduct tests using the company's Mythos model through an initiative called Project Glasswing. The official said the Mythos model identified certain vulnerabilities within hours, though it did not exploit them within that timeframe. Senator Mark Warner of Virginia referenced this testing during a June 11 hearing of the Senate Committee on Banking, Housing, and Urban Affairs. Warner said, "This tool broke into almost all of our classified systems, not in weeks but in hours." He attributed this information to General Joshua Rudd, who heads the National Security Agency and U.S. Cyber Command. The National Security Agency declined to comment, as did an Anthropic spokesman.

Earlier this month, the Trump administration issued a directive requiring Anthropic to prevent foreign nationals from using its latest artificial intelligence models, Fable 5 and Mythos 5. Anthropic revoked access to both models hours after the government's directive. Anthropic had previously stated in April that its Mythos AI model could pose a global cybersecurity threat due to its code-working capability, and it had tightly limited access to it due to cybersecurity concerns. The company provided access to Mythos to a small group of cybersecurity experts. Anthropic released Fable, a modified version of Mythos, to the public on Tuesday, June 9, stating it was safer for public use. On the Friday following June 9, the federal government informed Anthropic that Fable posed a threat to national security and placed export controls on it. Amazon CEO Andy Jassy also told government officials that Fable would be dangerous.

Anthropic disabled the models for all its customers to comply with the administration's directive. The company indicated it did not believe the government's actions were justified by the security concern it had raised. More than 100 cybersecurity experts and leaders from companies, including Adobe and Nvidia, sent a letter to the government asserting that Anthropic's Mythos models are effective at finding software flaws and weaponizing exploits. The letter also stated that it is dangerous to remove cyber defense capabilities "without a good reason" while adversaries are advancing.