Anthropic has launched an expanded version of its Cyber Verification Program, the company said in a statement, restructuring it into three access tiers and opening it to a broader range of qualifying security professionals. Each tier carries its own verification requirements and security controls, with access to its most capable models available across all tiers.
The revamped program brings together two separate earlier initiatives: Project Glasswing, which opened Claude Mythos access to organisations working to secure critical software, and the original program, which lowered safeguards on Claude Opus and Sonnet for vetted security teams. Both had been running for approximately six months before being merged. Anthropic's April release of Claude Mythos Preview generated concerns that AI could be used to compromise software before it had been secured.
From early April, roughly 50 initial partners used Claude Mythos Preview to scan codebases, and those partners identified over 10,000 high- or critical-severity flaws, Anthropic said in a separate statement. More than 33,000 vulnerabilities have been rated critical or high severity. Only organisations may apply for the Red Team tier, which extends access to authorised penetration testing and red teaming.
Security professionals, critical infrastructure operators, open-source maintainers and penetration testers can now apply for tiered access to Anthropic's most capable AI models with reduced safeguards, depending on the nature of their work and their organisation's ability to meet verification requirements.
Official statements: Anthropic
Reporting: iTnews
Written by Pick & Scroll News from the reporting and documents linked above.
