A US official has confirmed that Anthropic’s Mythos AI model uncovered vulnerabilities in some of the government’s most sensitive computer systems during a testing exercise, as reported by news agency AP. The report adds that the results of the testing exercise has raised fresh concerns about the power and risks of advanced artificial intelligence. The official speaking anonymously to the Associated Press, said the testing was conducted under Project Glasswing, an Anthropic-led initiative involving U.S. intelligence agencies and tech companies. Within hours, the Mythos model identified weaknesses in classified systems, though the official stressed this did not mean the AI exploited them.Democratic Sen. Mark Warner revealed during a June 11 Senate hearing that the tool “broke into almost all of our classified systems, not in weeks but in hours,” citing information from NSA and Cyber Command chief Gen. Joshua Rudd.
Tension with Trump administration
The report adds that despite the cooperation on testing, relations between AI giant Anthropic and the Trump administration remain strained. Earlier this month, the Trump administration ordered Anthropic to block foreign nationals from accessing its latest models, Fable 5 and Mythos 5, citing cybersecurity fears. Anthropic complied by disabling the models for all customers, but said the directive was unnecessary. The move followed Trump’s executive order requiring federal vetting of advanced AI systems for national security risks before public release.The report further adds that more than 100 cybersecurity executives and experts including leaders from Adobe and Nvidia urged the administration to lift restrictions, warning that sidelining Anthropic’s models could weaken U.S. defenses. They argued Mythos is effective at finding flaws but “not uniquely good” compared to other foundation and open-source models.
