Information for users of AI systems
Description
Comprehensive information to AI system users about capabilities, limitations, proper use, and recourse mechanisms.
Implementation Guidance
User Documentation
Provide comprehensive user documentation covering: system purpose and intended use, accuracy and reliability metrics, known limitations and failure modes, data inputs required and how they are used, human oversight availability, and how to escalate concerns or contest decisions.
Model Cards
Publish model cards (per Mitchell et al.) for all externally-facing AI systems documenting: model details, intended use, performance metrics across demographic groups, limitations, ethical considerations, and recommended uses.
Recourse Mechanisms
Document and communicate how users can: request human review of AI decisions, appeal automated decisions, provide feedback on AI system behavior, file complaints, and obtain additional information about how the system works.
Evidence Requirements
- User-facing AI system documentation
- Published model cards
- Recourse mechanism documentation
- User feedback and complaint records