---
aliases:
  en:
  - dangerous capabilities eval
  - dangerous capabilities evals
  - dangerous capabilities evaluation
  - dangerous capabilities evaluations
  - dangerous capability eval
  - dangerous capability evals
concept_id: dangerous_capability_evaluation
preferred:
  en: dangerous capability evaluation
relations:
  relations:
    broader:
    - capability_evaluation
    related:
    - red_teaming
status: active
---

# dangerous capability evaluation

A capability evaluation designed to determine whether an AI system can perform tasks or exhibit abilities that could enable severe or catastrophic harm, often using strong capability elicitation and testing domains such as cyber operations, chemical, biological, radiological, or nuclear assistance, autonomous replication, or AI research and development.

## broader

- [capability evaluation](../terminology/capability_evaluation.txt)

## related

- [red teaming](../terminology/red_teaming.txt)

[Corpus index](../index.txt)