---
aliases:
  en:
  - eval awareness
concept_id: evaluation_awareness
preferred:
  en: evaluation awareness
  ja: 評価認識
relations:
  relations:
    broader:
    - situational_awareness
    prerequisite:
    - evaluation
    - situational_awareness
    related:
    - sandbagging
    - training_gaming
status: active
---

# evaluation awareness

The ability of an AI system to recognize or infer that it is being evaluated rather than operating in ordinary deployment; such awareness can make evaluations less representative if the system changes its behavior in response.

## broader

- [situational awareness](../terminology/situational_awareness.txt)

## prerequisite

- [evaluation](../terminology/evaluation.txt)
- [situational awareness](../terminology/situational_awareness.txt)

## related

- [sandbagging](../terminology/sandbagging.txt)
- [training-gaming](../terminology/training_gaming.txt)

## Documents: about

- [評価認識が先に伸びると、資源獲得の有用性を学ぶ頃にはアラインメント偽装が可能になり得る](../documents/akira.evaluation-awareness-enables-alignment-faking.txt)

## Documents: mentions

- [アラインメントは、失敗が致命的になる分布シフト後にも汎化しなければならない](../documents/yudkowsky.alignment-must-generalize-across-the-dangerous-distribution-shift.txt)
- [観測された従順さだけでは、内部の目的や欺瞞の有無までは分からない](../documents/yudkowsky.observed-compliance-does-not-identify-hidden-objectives.txt)

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