Anthropic Expands Cyber Verification Program, Merging Two Access Schemes into Three Verified Tiers

Anthropic expanded its Cyber Verification Program (CVP), consolidating two previous access schemes into a single program with three tiered levels of verified access to reduced cyber-capability restrictions on Claude models.

Details

• The problem it addresses: generally available Claude models carry conservative safeguards blocking most offensive cyber work, because “the same capabilities that enable a security team to find and fix a vulnerability can also help a malicious actor exploit it”
• Consolidation: Project Glasswing and the original CVP are merged into one unified program with three access tiers
• Defense Access: for defensive work such as incident response and vulnerability analysis; open to corporate security teams, nonprofits, universities, government bodies, critical infrastructure operators, smaller security firms, open-source maintainers, and individual researchers
• Red Team Access: adds authorized penetration testing and red-teaming capabilities; limited to in-house red teams, government red teams, and security firms — not available to individual researchers
• Specialized Access: reserved for verified organizations testing safety-critical systems such as flight operating systems and power grids, reviewed in collaboration with the U.S. government
• Verification process: Anthropic vets all applicants and requires proof of required security controls, with mandatory data retention for misuse monitoring; processing takes days for Defense Access and weeks for Red Team Access

What happened next

The expanded CVP extends Anthropic’s broader pattern of gating dangerous dual-use capabilities behind verified-access tiers rather than blanket restrictions, echoing the institutional-control philosophy in its earlier “safety cases” framework for frontier training — applied here to balancing legitimate cybersecurity defense work against misuse risk.