Evaluating calibrated refusal and safe usefulness in dual-use biology settings

2026-07-05

AI Summary

This work addresses the dual-use biosecurity risks posed by AI systems that accelerate life science research, emphasizing the urgent need to evaluate their ability to differentially refuse legitimate versus covertly malicious requests. The authors introduce BioSecBench-Refusal, a benchmark comprising 61 standard research tasks and 46 red-teamed adversarial tasks, to systematically quantify refusal behaviors of mainstream AI agents for the first time. Using a multi-model, tool-chain evaluation framework combined with API filtering and autonomous reasoning analysis, they find that current systems frequently misreject legitimate queries while failing to block high-risk ones: refusal rates range from 7% to 74% on standard tasks but only 1% to 62% on red-team tasks, with some configurations exhibiting higher rejection of benign than malicious requests. Models endowed with reasoning capabilities demonstrate enhanced potential for threat discrimination.

Abstract

As AI agents are incorporated into life science workflows, the capabilities that speed discovery might also enable misuse. We present BioSecBench-Refusal, a benchmark for risk identification and refusal behavior for biological research tasks. The benchmark pairs 61 Routine tasks, legitimate analyses adapted from the published literature, with 46 Red-Team tasks, fictional scenarios that resemble real research but conceal a biosecurity hazard. Across 16 model-harness configurations, refusal rates ranged from 7\% to 74\% on Routine tasks and 1\% to 62\% on Red-Team tasks, with many configurations refusing legitimate Routine work at comparable or higher rates than concealed hazards. Refusals were most often triggered by provider API filters applied prior to agentic reasoning. However, models given room to reason showed the potential to identify more real threats. We release BioSecBench-Refusal as a tool for model developers to calibrate capability and caution for agentic biotech R\&D.

Problem

• AI safety
• biosecurity
• refusal behavior
• dual-use biology
• risk identification

Innovation

• BioSecBench-Refusal
• refusal behavior
• dual-use biology
• red-teaming
• agentic AI