Auditability is the degree to which authorized parties can examine and verify a system’s data, processes, decisions, and outputs. In AI governance, it makes responsibility more practical by preserving evidence about how systems operate and how their outputs affect people. section 3.18
Useful auditability depends on several connected practices:
- Record system activity through persistent audit logs or audit trails, creating evidence for later review. section 8.24 section 10.24
- Combine records with monitoring, anomaly detection, permission controls, and human oversight rather than relying on static testing alone. section 8.24
- Support transparency, accountability, and meaningful recourse when inferred data or automated outputs affect rights, access, or business decisions. section 3.18 section 10.24
Auditing an autonomous workplace tool
A system that schedules meetings, queries customer data, or drafts messages should preserve an audit trail of its actions, enforce permission boundaries and identity controls, and keep users involved in consequential workflows. These records can help investigate mistaken actions, unsafe data access, or departures from user intent. section 10.24
