Accountability for AI system decisions

Description

Clear assignment of accountability for AI system design, decisions, outcomes, and impacts to specific individuals or teams.

Implementation Guidance

Accountability Chain

Document an accountability chain for each AI system: who approved the use case, who validated the model, who authorized deployment, and who is accountable for ongoing performance and outcomes. Ensure at least one named individual (not a committee) holds ultimate accountability.

Decision Documentation

Require documented rationale for key AI decisions: model architecture selection, training data choices, fairness trade-offs, deployment scope, and risk acceptance. Link each decision to the accountable individual.

Consequences Framework

Define consequences for accountability failures: what happens when an AI system causes harm, how accountability is assessed, and how lessons learned feed back into governance improvements.

Evidence Requirements

• Accountability register mapping systems to individuals
• Decision logs with rationale and sign-off
• Incident reports tracing accountability chain
• Performance review records reflecting AI accountability