Human oversight
İnsan gözetimi
D6
Human oversight enables assigned people to understand relevant limits, monitor, interpret, disregard or override output, intervene, or stop AI-system use as appropriate and proportionate.
Review status: 2027-02-24
Technical explanation
For high-risk systems within the EU instrument's scope, oversight is designed around the system's risks, autonomy, use context, and the practical capabilities available during use.
Conceptual boundaries
Human oversight is not established by a person's presence alone and does not guarantee that errors will be caught.
Provider-neutral example
For an AI-supported triage queue, trained staff can see confidence and input limits, pause a case, override a recommendation, and route anomalies to a responsible owner.
Limitations
Enabled oversight capabilities do not guarantee that a person will identify every error.
Related concepts
Atomic claims and evidence
1.1Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text of 27 July 2026 — D6 human-role slice
- Source
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text of 27 July 2026 — D6 human-role slice
- Source role
- Authoritative source
- Exact locator
- Article 14(1)-(3), current consolidated text of 27 July 2026
- Supported claim
- For high-risk AI systems within its scope, Article 14 requires design and development enabling effective oversight by natural persons during use and states that oversight measures should be proportionate to risks, autonomy, and use context.
- Last verification
- Review due
- Scope limitation
- This is a legal requirement for the specified high-risk systems, not proof that every system with a human reviewer is effectively overseen.
1.2NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0) — D6 human and organizational slice
- Source
- NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0) — D6 human and organizational slice
- Source role
- Authoritative source
- Exact locator
- Appendix C, printed p. 41: human-AI configurations, challenge of outputs, and feedback
- Supported claim
- For high-risk AI systems within its scope, Article 14 requires design and development enabling effective oversight by natural persons during use and states that oversight measures should be proportionate to risks, autonomy, and use context.
- Last verification
- Review due
- Scope limitation
- This is a legal requirement for the specified high-risk systems, not proof that every system with a human reviewer is effectively overseen.
2.1Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text of 27 July 2026 — D6 human-role slice
- Source
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text of 27 July 2026 — D6 human-role slice
- Source role
- Authoritative source
- Exact locator
- Article 14(4)(a)-(e), current consolidated text of 27 July 2026
- Supported claim
- Article 14 lists understanding capacities and limitations, monitoring for anomalies, interpreting output, disregarding or overriding output, and intervening or stopping the system as capabilities to enable for assigned overseers, as appropriate and proportionate.
- Last verification
- Review due
- Scope limitation
- The list is tied to the cited high-risk-system provision and does not guarantee that an enabled person will identify every error.
2.2Parasuraman and Riley, Humans and Automation: Use, Misuse, Disuse, Abuse
- Source
- Parasuraman and Riley, Humans and Automation: Use, Misuse, Disuse, Abuse
- Source role
- Authoritative source
- Exact locator
- pp. 231-234: human use and misuse of automation
- Supported claim
- Article 14 lists understanding capacities and limitations, monitoring for anomalies, interpreting output, disregarding or overriding output, and intervening or stopping the system as capabilities to enable for assigned overseers, as appropriate and proportionate.
- Last verification
- Review due
- Scope limitation
- The list is tied to the cited high-risk-system provision and does not guarantee that an enabled person will identify every error.