Anthropic has merged its access programs for security defenders into three tiers and widened eligibility.
On October 6, Anthropic expanded its Cyber Verification Program (CVP). Generally available Claude models block most cyber work to prevent misuse, yet defenders need those limits eased. Anthropic therefore merged Project Glasswing and CVP into one program with three tiers: Defense Access (security operations, incident response, malware reverse-engineering, vulnerability analysis), Red Team Access (authorized penetration testing) and Specialized Access (safety-critical systems such as flight operating systems and power grids). Covered models are Claude Opus 5.5, Sonnet 5.5 and Mythos 5.1.
Each higher tier lifts blocks in practice. In Anthropic's test of 50 multi-stage cyber operation tasks, the Defense tier blocked 46 and the Red Team tier completed 34, matching unrestricted performance. In exchange, enrollment requires data retention so misuse can be monitored.
How many organizations clear the Red Team tier, whose review takes weeks, and whether defenders accept the data-retention condition. Those two points show whether the expansion turned into real use.