AI audit trails
Plan the records needed to examine an AI system's actions, data access, approvals, and outcomes.
View TN-ProtoWhat to record
- Actor. The person, agent, or service responsible for the action.
- Inputs. The sources and data available at the time.
- Authorization. The applicable permission, policy, or human approval.
- Outcome. The tool result, output, change, or exception.
Connect related events
Link records across the tools and services in a workflow so a reviewer can follow an action from the request through to the result.
Define the fields recorded in logs, who has access, and how long the records are retained. Apply access controls to the logs.
Implementation
Our TN-Proto research explores permissioned transaction records. See the logging notes for AI-assisted development and the guidance on privacy review records.