Article 14 — Human Oversight for High-Risk AI Systems

Legal Provision

Human oversight for high-risk AI systems: lays out measures for effective human oversight for high-risk AI systems to prevent/ minimise risks to health, safety, or fundamental rights.

Scope

Applies to entities who are providers and deployers of high-risk AI systems.

Requirement type

Duty: when providing or deploying a high-risk system, the entity must establish effective human oversight.

Actions to demonstrate compliance

A. Binary (yes/no)

• Built-in and documented oversight features (such as a stop button, alerts, explanations)
• Existence of guidance for safe use, the scope of the tool and how to detect issues, such as an instruction manual with oversight steps and automation bias warnings.
• Written oversight procedures and role descriptions.
• Documented training policies for staff responsible of conducting oversight
• Log of human intervention.

B. Non-binary (require a contextual assessment)

• Depth of risk analysis and rationale.
• Clarity & suitability of instructions.
• Appropriateness of oversight role design.
• Usability of alerts and interpretability tools.
• Relevance, effectiveness and quality of training.
• Completeness of pre-market tests.

Monitoring requirement

• Demonstrate oversight tools.
• Document guidance for safe use.
• Document training of responsible staff.
• Test logs verifying functionality of oversight mechanisms.
• Record of significant events
• Include human-oversight measures in the system’s overall risk assessment.. Review and update these measures when new risks or issues emerge.

Consequences of non-compliance

• Fines up to €15 million or 3% of worldwide annual turnover, whichever is higher.
• Other enforcement measures including warnings, suspensions or recalls that are put in place by Member States.