Anthropic finds 10k flaws with Mythos — and sells the fix
anthropic expanded project glasswing to 150 partners using mythos preview to find critical security flaws. it's selling protection via claude security while warning rival models could launch without safeguards within a year.
anthropic is scaling project glasswing to 150 partners in sectors including energy, water, and healthcare, using the claude mythos preview model to scan for security flaws. existing partners have identified over 10,000 vulnerabilities in systems protecting over 100 million people, with eu agency enisa reportedly in talks to join. anthropic plans to open mythos capabilities broadly soon but warns that competitors could release similar models within 6 to 12 months without adequate misuse safeguards. the initiative complements anthropic's commercial claude security product, which scans codebases for patches, effectively selling defense against the capabilities mythos demonstrates.