Orphan page — no incoming links
title: “CBRN AI Evaluations”
category: “ai”
created_at: “2026-10-13”
updated_at: “2026-10-17”
sources: [“2026-10-13-ai-models-hide-arbitrary-capabilities”]
Definition
Pre-deployment assessments focused on chemical, biological, radiological, and nuclear misuse pathways enabled by frontier models.
Key Points
- Often paired with capability-elicitation and red-team campaigns before public release.
- Policy gates may treat passing evals as necessary but not sufficient for deployment.