Anthropic Expands Claude Access for Vetted Cyber Teams as Glasswing Finds 129,000 Flaws

Anthropic has broadened its Cyber Verification Program, letting vetted security professionals use Claude models—including Opus 5.5, Sonnet 5.5 and Mythos 5.1—with fewer built-in restrictions for defensive and authorized offensive security work. The move follows results from Project Glasswing, which Anthropic says identified roughly 129,000 verified software vulnerabilities between April and July 2026, plus another 5,500 found via open-source scanning through October, with over 33,000 rated critical or high severity. The program now offers three tiers—Defense Access, Red Team Access, and a highly restricted Specialized Access—each unlocking different levels of safeguard relaxation for incident response, malware analysis, and penetration testing. Separately, outside research noted that only a small fraction of flaws surfaced by Anthropic-linked efforts have actually been exploited in the wild, underscoring that AI-assisted discovery doesn't always equate to real-world risk.