Measuring Biological Capabilities and Risks of AI Agents

2026-06-18

AI Summary

Current applications of AI agents in biology lack a systematic evaluation framework, and their risk assessments are often confounded by implicit design choices. This work proposes the first explanation-sensitive evaluation framework tailored to biological AI agents, integrating methods from AI evaluation, biosafety, and policy research through an interdisciplinary approach. It systematically examines how decisions in defining, constructing, executing, scoring, and documenting evaluations influence risk inference. Drawing on empirical analysis, the study identifies a set of design considerations that culminate in an actionable evaluation guideline. This framework enhances the transparency and interpretability of AI-driven biological risk assessments, offering practical guidance for policymakers, funding agencies, and biosafety practitioners to inform high-impact research and systematic oversight.

Abstract

This paper addresses a rapidly emerging policy challenge: how to generate and interpret credible evidence about the biological capabilities and risks of AI scientists, or agentic AI systems capable of autonomously or collaboratively performing multi-step scientific tasks. As these systems enter real research workflows, decision-makers increasingly face evaluation results whose meaning depends on underlying design choices that are often implicit or under-documented. We synthesize current evidence on AI-enabled biological risks and introduce biological agentic evaluations as a promising, but interpretation-sensitive, tool for assessing these systems. Our central contribution is a set of practical, experience-grounded considerations -- drawing from our own evaluations -- that show how choices around defining, designing, running, scoring, and documenting evaluations materially shape what results do and do not imply about risk. The analysis is intended to help policymakers interpret biological evaluation outputs with appropriate caution; guide public and private funders toward high-leverage investments in AI-biology evaluation research; and support biosecurity practitioners assessing emerging AI systems. A secondary audience includes researchers designing or conducting agentic evaluations within frontier AI labs, AI providers, scientific institutions, and third-party evaluation organizations.

Problem

• AI agents
• biological risks
• agentic evaluations
• scientific tasks
• evaluation interpretation

Innovation

• biological agentic evaluations
• AI risk assessment
• evaluation design
• AI biosecurity
• agentic AI systems