Anthropic AI Finds Flaws In Classified U.S. Systems Within Hours
By CU Today Staff —
WASHINGTON--Anthropic’s AI model Mythos identified vulnerabilities in highly sensitive U.S. government computer systems during a classified testing exercise conducted with intelligence agencies, according to reporting by Reuters.
Reuters reported that Anthropic partnered with U.S. intelligence agencies under a restricted initiative known as Project Glasswing, which was designed to uncover and remediate weaknesses in critical government software before they could be exploited by adversaries. The testing reportedly involved some of the government's most sensitive systems.
During a recent congressional hearing, Sen. Mark Warner said he had been informed by National Security Agency Director Joshua Rudd that Mythos "broke into almost all of our classified systems, not in weeks, but in hours," according to Reuters. However, the Associated Press, citing a U.S. official, noted that while the model identified vulnerabilities within hours, that did not necessarily mean it successfully exploited those weaknesses in the same timeframe.
Reuters also reported that Anthropic's relationship with the federal government has become increasingly strained. The company previously restricted military use of its AI models for domestic surveillance and fully autonomous weapons systems, and this month the government ordered Anthropic to suspend exports of its latest Mythos and Fable models worldwide to foreign nationals over national security concerns. Reuters noted that National Security Agency reportedly lost access to Mythos amid the dispute, according to earlier reporting by The New York Times.
Originally reported by CU Today.