Anthropic’s Mythos artificial intelligence model identified vulnerabilities in highly sensitive and secure U.S. government computer systems during a testing exercise conducted in partnership with U.S. intelligence agencies, a U.S. official told The Associated Press on Tuesday.
The official, who spoke on condition of anonymity to discuss the matter, said the testing was carried out through an Anthropic initiative called Project Glasswing. That project brought together technology companies and other firms in an effort to secure the world’s critical software from what the official described as “severe” fallout that the Mythos model could pose to public safety, national security and the economy.
The model identified certain vulnerabilities within hours, according to the official, but that does not mean the model was able to exploit them within that time frame, the official added.
The tests come amid a broader push by the Trump administration to restrict access to advanced AI models over national security concerns. The White House earlier this year ordered all federal agencies to stop using Anthropic’s AI, and the Pentagon has labeled the company a supply-chain risk — a designation a federal judge later blocked. Anthropic has also held talks with administration officials seeking to restore access to its most powerful models.
The specific vulnerabilities identified during the Mythos exercise were not disclosed. The official did not say whether the government has since patched the flaws.