2026年10月6日，Anthropic宣布对其网络验证计划（CVP）进行全面升级。
该计划允许经过审查的网络安全专业人员在降低安全限制的情况下，使用其先进的AI模型，包括Claude Opus 5.5、Sonnet 5.5和Mythos 5.1。
Anthropic Expands Cyber Verification Program After AI Models Detect Over 134,000 Software Vulnerabilities On October 6, 2026, Anthropic announced a revamped Cyber Verification Program (CVP) that grants vetted cybersecurity professionals access to its advanced AI models—including Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1—with reduced safeguards.
此次扩展源于Anthropic及其“玻璃翼”（Project Glasswing）计划的发现，该计划在2026年4月至10月期间共识别出超过134,500个已验证的软件漏洞，其中超过33,000个被评为严重或高危级别。Anthropic指出，实际漏洞总数可能被低估。
The expansion follows findings from Anthropic and its Project Glasswing initiative, which together identified more than 134,500 verified software vulnerabilities between April and October 2026, with over 33,000 rated as critical or high severity. Anthropic noted that total vulnerabilities are likely undercounted.
重组后的框架分为三个层级：防御层、红队层，以及与美国政府共同审查的专业层，旨在评估电网、航空控制和银行间交易基础设施等关键安全系统。
The restructured framework features three tiers: Defense, Red Team, and a Specialized tier vetted alongside the U.S. government to evaluate safety-critical systems such as power grids, aviation controls, and interbank transaction infrastructure.