---
aliases:
  en:
  - AI control protocol
  - control protocols
concept_id: control_protocol
preferred:
  en: control protocol
relations:
  relations:
    broader:
    - ai_control
    prerequisite:
    - ai_control
status: active
---

# control protocol

A specified procedure or system architecture for using potentially untrusted AI systems while limiting unacceptable outcomes, combining model roles with safety measures such as monitoring, auditing, deferral, editing, resampling, or other constraints so useful work can be obtained even if some models intentionally try to subvert the measures.

## broader

- [AI control](../terminology/ai_control.txt)

## prerequisite

- [AI control](../terminology/ai_control.txt)

## Documents: mentions

- [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)