Anthropic’s newest model excels at finding security vulnerabilities—but raises fresh cybersecurity risks — Blankdot