---
aliases:
  en:
  - capability thresholds
concept_id: capability_threshold
preferred:
  en: capability threshold
relations:
  relations:
    prerequisite:
    - capability
    related:
    - dangerous_capability_evaluation
status: active
---

# capability threshold

A specified level of performance in an AI capability domain that a safety or governance framework treats as a decision boundary, such that reaching or plausibly exceeding it triggers stronger evaluation, safeguards, or changes to development or deployment decisions.

## prerequisite

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

## related

- [dangerous capability evaluation](../terminology/dangerous_capability_evaluation.txt)

## Documents: mentions

- [アラインメントは最初の危険な試行で成功する必要がある](../documents/yudkowsky.first-critical-try.txt)

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