Anthropic has restructured its Cyber Verification Program into a three-tier access system for security teams. The new framework integrates the former Project Glasswing, which has been active since April. The program currently supports Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1, with plans to include future models upon release.
The entry-level Defense Access tier covers incident response and malware analysis for companies, universities, and government bodies. The Red Team Access tier enables authorized penetration testing, while the Specialized Access tier is reserved for organizations securing critical infrastructure like power grids and telecom networks, requiring deep vetting with the U.S. government.
To participate, organizations must consent to data retention for misuse monitoring. Later this year, Anthropic will introduce Enterprise Frontier Safeguards, allowing customers to store this data within their own cloud infrastructure. The program is accessible via the Claude Platform, Google Vertex AI, Microsoft Foundry, and Amazon Bedrock.
Internal testing on the CyScenarioBench benchmark showed that while standard models block most offensive tasks, the Red Team tier allowed models to complete 34 out of 50 runs. Data from the integrated Glasswing initiative revealed that partners identified at least 129,000 verified vulnerabilities between April and July, with over 33,000 classified as critical or high severity.
