Human Oversight
The roles and mechanisms through which people monitor, check, approve, return for revision, intervene in, stop, or override operations or decisions involving AI systems.
Example
Before a high-impact action is executed, a person reviews the proposed action and either approves it or sends it back for revision.
Distinctions and cautions
The presence of Human Oversight does not guarantee safety or correctness. Its effectiveness depends on when intervention is possible, the supervisor's knowledge and attention, workload, information presented in the interface, and authority to stop or return a process for revision. It is not the same as human scoring of evaluation examples or merely avoiding overreliance.
Evidence
View evidence →
AI Risk Management Framework 1.0
- Title
- AI Risk Management Framework 1.0
- Author / organization
- NIST
- Year
- 2023
- Source type / publication status
- Government official technical publication; published
- Supported concepts / claim context
- Human Oversight, roles, challenge/override context.
- Scope and limitations
- Broad risk framework, not an experimental estimate of effect size.
- Access / version note
- Published official source.
- Recorded official URL / DOI / identifier
- DOI:
10.6028/NIST.AI.100-1(DOI resolver)
Artificial Intelligence Risk Management Framework: Generative AI Profile
- Title
- Artificial Intelligence Risk Management Framework: Generative AI Profile
- Author / organization
- NIST
- Year
- 2024
- Source type / publication status
- Government official technical publication; published
- Supported concepts / claim context
- Overreliance, automation bias, oversight.
- Scope and limitations
- Contextual risk-management guidance for GenAI.
- Access / version note
- Published official source.
- Recorded official URL / DOI / identifier
- DOI:
10.6028/NIST.AI.600-1(DOI resolver)