Cyber Verification Program
- Source
- anthropic.com
- Author
- Anthropic editorial sitemap
- Date
Frontier models often refuse legitimate defensive security work, and labs have been testing ways to grant exceptions to vetted users. Anthropic folded Project Glasswing and the Cyber Verification Program into a single scheme with three access tiers. The Defense and Red Team tiers relax cyber safeguards on Anthropic's top models for authorized penetration testing and related work. Review times differ: Defense Access takes days, while Red Team Access takes weeks. Some actions remain blocked at every tier, including deploying ransomware and actions that risk physical harm or mass disruption. For security teams, the practical step is to apply before a project needs the access, given the review lead time. For everyone else, it shows that access to stronger cyber capability is moving toward identity verification rather than one public policy.

Context
Anthropic's expanded Cyber Verification Program, announced October 6, 2026, ties access to advanced cybersecurity capabilities to the scope of an applicant's work. Defense Access covers activities such as incident response and vulnerability analysis. Red Team Access adds authorized , while Specialized Access is reserved for deeply vetted organizations testing sensitive systems. The distinction matters because security work can require techniques that ordinary model safeguards block. Verification provides a route to broader capabilities with tier-specific controls; Red Team Access is limited to organizations and still restricts actions that could cause physical harm or mass disruption. Access is not automatic: Anthropic says it verifies applicants, requests proof of required security controls, and requires data retention so it can monitor for misuse.
- red teaming — Deliberately attacking your own AI system to find what makes it fail before someone else does.
Checking sign-in…
Loading comments…






