Accountability requires responsible parties to explain, justify, and demonstrate compliance for an AI or ML system’s decisions and effects. It makes data protection, risk mitigation, design choices, and compliance evidence visible rather than assumed. The customer must be able to take responsibility for compliance, while the team identifies controller and processor relationships across the AI supply chain. section 2.6
Turn accountability into requirements and production checkpoints:
- Document how the system complies with relevant obligations and justify significant choices. section 2.6
- Identify risks, assess their potential effects, and record mitigation actions, especially for marginalized or disempowered groups. section 2.6
- Assign people with the authority and competence to judge the system’s utility when it may cause harm. section 9.7
- Establish governance that remains effective as personnel change, and measure performance through effects on users and others in the operating context. section 9.7
- Check that interfaces and controls help users understand and manage system behavior, and consult people affected by deployment. section 9.7
Accountability checkpoint
Before deployment, require an evidence pack containing compliance documentation, controller/processor responsibilities, assessed harms, mitigation decisions, governance ownership, user-impact measures, interface explanations, and control testing. Escalate unresolved harm to an authorized decision-maker. section 2.6 section 9.7
