These findings follow the expansion of the Cyber Verification Program into a tiered system of Defense, Red Team, and Specialized access. The program provides vetted professionals with Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1, the latter of which shares its architecture with the public Fable 5.1 but operates with permissive safeguards for authorized penetration testing and offensive security operations.
Tuesday, October 20, 2026
Story
Anthropic AI Finds 129,000 Vulnerabilities via New Red Team Tier
Cybersecurity14 days ago
Security partners using a permissive AI model identified over 33,000 high-severity flaws in recent tests of the company's latest security offerings. Project Glasswing partners discovered at least 129,000 verified vulnerabilities in total, an impact that Anthropic suggests could be five times higher. The system's Red Team Opus 5.5 model completed 34 of 50 tasks with no blocks, matching the performance of a version of the model with all safeguards removed.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.