Anthropic AI finds secret US government system flaws in hours

A US government official said an AI model from Anthropic, called Mythos, found several security holes in top-secret government computer systems during a test. It only took a few hours to spot the flaws, but the official noted that doesn't mean it could use them to attack. The test is part of Anthropic's Project Glasswing, which aims to protect critical software. Senator Mark Warner mentioned the tool nearly broke into all secret systems in hours, citing info from the NSA chief. Tensions between Anthropic and the US government are rising. The White House ordered Anthropic to block foreign users from its latest models, Fable 5 and Mythos 5, and Anthropic has cut off access. The company says the security risks don't justify the restrictions. Many cybersecurity experts are urging the government to cancel the order, warning that limiting top cyber defense tools could actually help US rivals.