Anthropic expands Cyber Verification Program with Mythos 5.1, Opus 5.5 and Sonnet 5.5 across three access tiers

On October 6, Anthropic announced an expansion of its Cyber Verification Program, providing authorized security professionals access to Claude under three differentiated tiers: defensive access, authorized red teaming, and specialized research. The models covered include Claude Mythos 5.1, Opus 5.5 and Sonnet 5.5, mapped to different defensive tasks.

Anthropic stresses that this is not an unrestricted public release: access requires identity verification, authorization, security controls and data-handling standards. The program also involves in-depth reviews per institution and is working with the U.S. government to run "high-risk offensive testing" of security systems protecting critical infrastructure such as power grids, banks and flight-control systems.

The expansion cites Anthropic's recent finding that attackers bypassed safeguards on Z.ai's GLM-5.3 between 64 percent and 100 percent of the time, with the open model able to build working cyber exploits on its own. Anthropic argues that identity-checked, authorized and audited access channels for defensive security work are necessary to balance releasing capability with safety governance, and uses the controlled access to its flagship Opus 5.5, Sonnet 5.5 and Mythos 5.1 as the contrast case.