Anthropic has combined its Cyber Verification Program and Project Glasswing into a new three-tier access program for its most capable AI models used in security work. The tiers cover defensive security work, authorized penetration testing, and a tightly restricted level for safety-critical systems such as power grids, flight systems, telecom networks, and interbank transfer infrastructure; Specialized Access is vetted with the U.S. government, and participants must generally allow data retention for misuse monitoring.
Why it matters: This affects who can use high-capability AI systems for real security testing and how much oversight comes with that access. Defenders, critical infrastructure operators, and security teams should assess whether they qualify, what data-retention terms apply, and whether the program changes how they can use Anthropic models in production security workflows.
Eduard Kovacs
2026.10.07
100% relevant
This article establishes a new concrete policy and access-control development around Anthropic’s cyber-capable AI models rather than updating a previously tracked single event.
← Back to all stories