Anthropic's Private Mythos Model Found Flaws in Classified US Systems Within Hours

According to a report by the Associated Press, Anthropic's non-public AI model "Mythos" identified vulnerabilities in classified US government systems within hours during a joint testing exercise with US intelligence agencies. Citing an anonymous US official, the report says Anthropic worked with intelligence agencies through its Project Glasswing initiative, and that Mythos discovered the flaws within "hours." The model did not actually exploit the systems, however. CNBC and Reuters reported on the matter on the same day and the day after.