AI Glossary

AI audit

AI system audit, AI review

An AI audit is a structured review of an AI system: its data, behavior, risks and compliance with the rules in place. It checks whether the model does what was claimed and produces evidence for oversight purposes.

It examines the model's data, behavior and decisions against the defined requirements and risks.
It ends in evidence: records, test results and a list of the gaps found.
It's repeated periodically, because the model and the data change over time.

An AI audit is a systematic check of whether a system works as intended and stays within the rules in place. It reviews the input data, how the model behaves, the decisions it makes, and the risks — for example, bias in the outputs or the processing of sensitive data. The aim is an honest picture of the current state, not a declaration.

An audit differs from model evaluation in its broader scope: it covers the process, the roles and compliance, not just the accuracy of the answers. It gives AI governance evidence that the rules are being followed. Because the system and the data change, an audit is repeated periodically rather than done once.

Related terms

In guides