Anthropic is expanding its Cyber Verification Program to provide vetted security professionals with tiered access to advanced AI models for identifying and patching critical software vulnerabilities.
Key Points
- Anthropic’s Project Glasswing identified over 134,500 verified software vulnerabilities between April and October 2026.
- The new Cyber Verification Program offers three access tiers—Defense, Red Team, and Specialized—for testing models like Claude Opus 5.5 and Claude Sonnet 5.5.
- Red Team Access allows researchers to bypass standard AI safeguards to conduct authorized penetration testing and vulnerability validation.
- Only two of the 300 vulnerabilities identified by the initiative have been exploited in the wild, according to VulnCheck research.
- Veracode reports that 44% of AI-generated code tasks introduce new security risks, highlighting ongoing challenges with automated patching.