Anthropic Introduces 3-Tier Cyber Verification Program for AI Access
Anthropic launched a three-tier program controlling defender access to less-restricted Claude cyber models.
Anthropic combined its Cyber Verification Program and Project Glasswing into three tiers of access to models with fewer cyber blocks, including Claude Opus 5.5, Sonnet 5.5, Fable 5.1, and Mythos 5.1. Defense Access covers SOC, malware analysis, and vulnerability work; Red Team Access adds authorized testing for organizations only; Specialized Access, vetted with the US government, is for safety-critical systems such as grids and telecom. Glasswing partners verified at least 129,000 vulnerabilities from April to July, more than 33,000 critical or high, and Anthropic scanning found 5,500 more. Members must allow data retention for misuse monitoring; CVP is on the Claude Platform, Vertex AI, and Microsoft Foundry, with limited Amazon Bedrock access.