Anthropic’s expanded Cyber Verification Program gives vetted security professionals access to Claude models with fewer restrictions. Its predecessor’s partners reported at least 129,000 confirmed vulnerabilities from April to July 2026, including over 33,000 rated high-severity or critical.