Claude Opus 5 Finds Software Vulnerabilities While Blocking Exploit Generation

A summary of Anthropic's Claude Opus 5: the report describes Opus 5 as a defensive-focused, highly aligned large AI model that can identify software flaws while incorporating layered safeguards (cyber classifiers, defaulting to restricted versions, and a Cyber Verification Program) to block exploit generation and offensive workflows. It highlights Opus 5's strong code-reasoning and vulnerability-finding capabilities relative to other models, and includes a brief, unelaborated alert line claiming "20+ government sites delivered malware," without providing technical details or indicators.