Description
Systematic design process incorporating safety, security, fairness, transparency, and accountability by design from inception.
Implementation Guidance
Design Principles
Embed responsible AI principles at the design phase: define fairness constraints before model selection, specify explainability requirements based on use case risk, incorporate privacy-by-design (data minimization, purpose limitation), and plan for human oversight interfaces.
Design Documentation
Produce a system design specification covering: problem definition, intended use and users, input/output specifications, performance requirements, fairness criteria, safety boundaries, and failure modes analysis. Require design review approval before development begins.
Risk-Informed Design
Conduct a design-phase risk assessment identifying potential harms, bias vectors, security attack surfaces, and failure scenarios. Document design decisions made to mitigate identified risks and the rationale for trade-offs.
Evidence Requirements
- AI system design specifications
- Design review approval records
- Design-phase risk assessments
- Responsible AI requirements per system