---
aliases:
  en:
  - AI alignment failure
concept_id: alignment_failure
preferred:
  en: alignment failure
  ja: アラインメント失敗
relations:
  relations:
    broader:
    - misalignment
    prerequisite:
    - ai_alignment
    - misalignment
    related:
    - alignment_risk
status: active
---

# alignment failure

An instance or failure mode in which an AI system's specified objective, learned goals, or resulting behavior diverges materially from what its designers intended; alignment failures can arise through outer misalignment, inner misalignment, or other mismatches between the system and its intended target.

## broader

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

## prerequisite

- [AI alignment](../terminology/ai_alignment.txt)
- [misalignment](../terminology/misalignment.txt)

## related

- [alignment risk](../terminology/alignment_risk.txt)

## Documents: about

- [アラインメント研究の自動化は、研究AIが意図的に妨害しなくても失敗し得る](../documents/bowkis_et_al.automated-alignment-can-fail-without-deliberate-sabotage.txt)

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