Anthropic's Mythos Model Exposes Vulnerabilities in US Classified Systems

Anthropic's Mythos model identified vulnerabilities in highly sensitive U.S. government computer systems during a testing exercise. The company collaborated with Washington's intelligence agencies to conduct these tests under Project Glasswing, a restricted program aimed at finding and fixing vulnerabilities in critical software before they can be exploited by attackers.

Senator Mark Warner of Virginia referred to the testing in a congressional hearing this month, stating that National Security Agency chief Joshua Rudd informed him that Mythos "broke into almost all of our classified systems, not in weeks, but in hours." An unidentified U.S. official noted that while Mythos identified certain vulnerabilities quickly, it did not necessarily mean the model could exploit them within that same timeframe.

The White House, Anthropic, and the Department of Defense did not respond to requests for comment. Anthropic's relationship with the U.S. government has been complicated; the company previously refused to allow the military to use its AI models for domestic surveillance and fully autonomous weapons systems, leading to its placement on a national security blacklist. Additionally, the U.S. government recently ordered Anthropic to suspend exports of its latest Mythos and Fable AI models to all foreign destinations, citing national security concerns. Reports indicated that the NSA lost access to Mythos amid this ongoing dispute.