Human Oversight
Requirement that humans maintain meaningful oversight over AI-assisted decisions, including the ability to understand, override, and intervene in automated outputs.
What Counts
- Meaningful human review with documented criteria
- Override and intervention capabilities
- Training requirements for human reviewers
- Prohibition on defaulting to AI system output
- Requirements that reviewers understand system limitations
What Does Not Count
- Rubber-stamping AI outputs without genuine review
- Human involvement that lacks authority to override
- Automated appeals processes without human involvement
Related Terms
- Human-in-the-loop
- Human control
- Meaningful human review
- Override capability
Implementing Regulations
| Regulation | Jurisdiction | Status | Provisions |
|---|---|---|---|
| California SB 1120 (Physicians Make Decisions Act) | California | enforcing | 1 |
| CMS Medicare Advantage AI Rule | United States | enforcing | 1 |
| Colorado ADMT (SB 24-205) | Colorado | pending replacement | 1 |
| Connecticut SB 1295 | Connecticut | enacted | 1 |
| EU AI Act | European Union | phased enforcement | 1 |