Testing Large Language Model Agents on the Use of Biological Tools for Nucleic Acid Synthesis Screening Evasion

RAND's Center on AI, Security, and Technology published a new research report evaluating whether large language model (LLM) agents can exploit biological tools (BTs) to redesign peptides and proteins in ways that evade nucleic acid synthesis screening — a key biosecurity safeguard. The authors find that 'LLM agents demonstrate emerging ability to use BTs for nucleic acid synthesis screening evasion, although success is inconsistent and model safeguards frequently limit testing of closed-weight systems.' This is a peer-reviewed RAND research report (published Aug 11, 2026) building on RAND's ongoing biosecurity threat-model workstream, following up on the June 2026 initial assessment 'Can LLM Agents Select and Engage with Biological Tools?'. The report assesses a specific threat model — agentic misuse lowering expertise barriers for hazardous biodesign — rather than general capability benchmarking.

This is a direct, empirical signal that agentic AI systems are beginning to demonstrate capabilities relevant to circumventing biosecurity screening infrastructure, informing frontier lab safeguard design, national biosecurity policy, and dual-use research oversight decisions.

Frontier labs and biosecurity policy teams should review closed-weight safeguard efficacy against agentic misuse and brief risk committees on nucleic acid synthesis screening gaps.