A Threshold Exceedance Framework for CBRN Uplift Evaluation in Frontier Language Models

The paper introduces Threshold Exceedance Criteria (TEC), a framework for making CBRN uplift evaluations more comparable by separating participant eligibility, threat scope, and statistical estimation of material uplift. It distinguishes generative uplift, where a model helps create a plan from scratch, from revisionist uplift, where it refines an existing plan. In a large controlled pre-release study, some model-assisted plans received expert-equivalent instructional ratings, but confirmed material uplift was limited to the radiological domain. The authors stress that these results informed mitigation and governance decisions rather than describing deployed-model behavior.