Measuring Biological Capabilities and Risks of AI Agents
This perspective examines biological agentic evaluations as a tool for assessing risks from increasingly autonomous AI systems and offers practical guidance on how evaluation design shapes conclusions about biological capability and risk