---
aliases:
  en:
  - AI control safety case
  - control safety cases
concept_id: control_safety_case
preferred:
  en: control safety case
relations:
  relations:
    broader:
    - safety_case
    prerequisite:
    - capability_elicitation
    - control_evaluation
    - control_protocol
status: active
---

# control safety case

A structured, evidence-based argument that specified control measures make an AI deployment sufficiently safe despite adversarial model behavior, typically relying on control evaluations together with claims that dangerous capabilities were adequately elicited, the evaluation conservatively represents deployment, and the resulting risk is extrapolated appropriately.

## broader

- [safety case](../terminology/safety_case.txt)

## prerequisite

- [capability elicitation](../terminology/capability_elicitation.txt)
- [control evaluation](../terminology/control_evaluation.txt)
- [control protocol](../terminology/control_protocol.txt)

## Documents: about

- [control safety caseには、能力の十分な引き出しと、評価・外挿の保守性が必要になる](../documents/korbak_et_al.control-safety-cases-depend-on-elicitation-transfer-and-extrapolation.txt)

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