Anthropic’s testing reveals classified U.S. government systems have vulnerabilities, says anonymous source

An artificial intelligence model developed by Anthropic identified vulnerabilities in classified U.S. government computer systems within hours during a testing exercise, a U.S. official told The Associated Press on Tuesday. The official, who spoke anonymously, stated that Anthropic collaborated with U.S. intelligence agencies to conduct these tests using its Mythos model. While the model detected vulnerabilities rapidly, the official clarified that this does not imply it could exploit them within the same timeframe. This testing was part of Anthropic's Project Glasswing initiative, which aims to secure critical software from potential severe impacts on public safety, national security, and the economy. Senator Mark Warner of Virginia had previously referenced this testing during a June 11 hearing, attributing the information to General Joshua Rudd, head of the National Security Agency and U.S. Cyber Command, stating the tool "broke into almost all of our classified systems, not in weeks but in hours." The NSA and Anthropic declined to comment on the matter. This collaboration occurs amidst growing tensions between Anthropic and the Trump administration, with the company expressing concerns about potential military applications of its AI and the administration imposing restrictions on its models, including a directive requiring Anthropic to prevent foreign nationals from accessing its latest models, Fable 5 and Mythos 5.