How do I generate compliance artifacts for AI agents?

Answer

Six document types form a complete compliance evidence package:

1. System Card — describes the agent's purpose, architecture, deployment context, and evaluation results.
2. Model Card — details the underlying model's capabilities, limitations, and bias characteristics.
3. Data Processing Agreement — documents data flows, processing purposes, and controller/processor relationships.
4. Privacy Policy — specifies what data the agent processes and user rights.
5. Safety Case — structured argument with evidence that the agent is acceptably safe.
6. Transparency Report — periodic aggregate of all compliance metrics. Generate these from live platform data, not manually.