Glossary term
Glossary term
Safety and Alignment
Risk that AI could assist chemical, biological, radiological, or nuclear harm. Frontier AI governance treats CBRN enablement as a severe-risk category requiring specialized evaluation and access controls. Most organizations will not evaluate CBRN risk directly, but they should know when a vendor, model, or use case requires specialist safety evidence.
OpenAI's Preparedness Framework explicitly defines CBRN as a risk category with Critical capability thresholds.
Anthropic, OpenAI, and Google DeepMind have public commitments under the Seoul Frontier AI Safety Commitments to evaluate and mitigate CBRN risks.
The Bipartisan AI Insight Forums (US Senate, 2023 to 2024) and the UK AISI focus specifically on CBRN risk uplift from frontier models.