5 min read
News Brief Anthropic's Claude Opus 5 system card documents the model's alignment improvements, OSS-Fuzz benchmark results, and intentional limits on offensive cybersecurity capability. The model blocks binary vulnerability scanning, pen testing assistance, and exploit generation, with flagged requests falling back to an older Claude version rather than failing outright.