AI Audit Trail

An AI audit trail is a tamper-resistant record of what an AI system did and why: the inputs it received, the knowledge it retrieved, the rules and tools it invoked, the reasoning behind its output, and any human approval or override along the way. It exists so that a compliance team, an auditor, or a regulator can reconstruct a specific decision after the fact.

The distinction that matters is between a log and a decision record. A log confirms that a workflow ran at a given time. A decision record explains why the system compiled the response it did, which agents or models it called, what they changed, and which alternatives were considered. Under the EU AI Act, both providers and deployers of high-risk systems carry logging obligations, so the audit trail is a legal requirement as well as a trust mechanism. For predictive models, the trail also covers lineage: which model version, trained on which data, produced the output. Darwinbox versions every model with lineage and hyperparameters, logs agent reasoning steps and source inputs, and in Cortex frames this as a receipt attached to every action.