Anthropic’s Mythos model identified vulnerabilities in highly sensitive US government computer systems during a testing exercise, the Associated Press has reported.
Anthropic teamed up with Washington’s intelligence agencies to conduct tests using Mythos under Project Glasswing, the AP said, referring to a restricted programme designed to find and fix vulnerabilities in critical software before attackers could exploit them.
Senator Mark Warner of Virginia referred to the testing in a congressional hearing this month, saying he had been informed by National Security Agency chief Joshua Rudd that Mythos “broke into almost all of our classified systems, not in weeks, but in hours.”
Citing an unidentified US official, the AP said that although Mythos identified certain vulnerabilities within hours, that did not mean the model was able to exploit them within that time.
IPO-bound Anthropic’s relationship with the US government has been rocky.
The company refused to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems and the government retaliated by putting it on a national security blacklist.
The US government also this month ordered the company to suspend exports of its latest Mythos and Fable AI models to destinations worldwide and all foreign nationals, citing national security concerns.
The New York Times reported earlier in the day that the NSA lost access to Mythos amid the dispute.
