Abstract
Section 7.2 of the Claude Opus 4 and Claude Sonnet 4 system card presents Anthropic’s Chemical, Biological, Radiological, and Nuclear (CBRN) evaluations conducted under the Responsible Scaling Policy. The section describes ASL-3 and ASL-4 threat models focused on biological risks, and reports results from bioweapons acquisition uplift trials, expert red-teaming, long-form virology tasks, multimodal virology, bioweapons knowledge questions, DNA synthesis screening evasion, LAB-Bench, creative biology, short-horizon computational biology tasks, and ASL-4 expert red-teaming. Based on these evaluations, Anthropic was unable to rule out ASL-3 uplift for Claude Opus 4 and deployed it under ASL-3 safeguards, while Claude Sonnet 4 was deployed under ASL-2.