An AI impact assessment examines the potential effects of an AI system on people and society before deployment.
When to Do One
- Systems affecting people's rights, opportunities or safety.
- Use of sensitive personal data.
- Public-facing systems at scale.
- Where regulations require assessments.
What It Covers
- Purpose: what the system does and why.
- Stakeholders: who's affected, including indirectly.
- Benefits and harms: accuracy, fairness, privacy, safety, autonomy.
- Data: sources, quality and consent.
- Mitigations: controls to reduce risks.
- Oversight: human review and monitoring.
- Residual risk: what remains and who accepts it.
Involve Diverse Voices
Include people affected by the system, domain experts, legal and ethics specialists.
Proportionality
Scale the assessment to the risk. Low-risk internal tools need a light review.
Living Document
Update the assessment when the system, data or use changes, and review it after incidents.
Related Assessments
Coordinate with privacy impact assessments and security reviews to avoid duplication.