According to @CNBC, Anthropic’s Mythos model identified vulnerabilities in classified U.S. systems, signaling new AI security testing uses, per AP reporting.