What it means
An audit trail is a secure, time-stamped record of what happened in a system: who did what, when, with which inputs and outputs, and which decisions were made. In AI systems, it allows actions and outputs to be reviewed and traced later.
Real-world example
After a citizen complains that an AI-assisted notice from a city office contained incorrect information, the office uses its audit trail to see which documents the system retrieved, what draft it generated, which caseworker reviewed it, and what was changed before sending. This shows the error came from an outdated policy document, which is then corrected.
Why it matters for AI adoption
Audit trails support accountability, error correction, complaints handling, security investigations and regulatory requirements; the EU AI Act, for example, requires logging capabilities for high-risk systems. Logging must also respect data protection: it should record what is needed, be protected and be deleted when no longer required.
Questions to ask
- What does the system log, and can we reconstruct how a specific output was produced?
- Who can access the logs, and how long are they kept?