Data
- List which personal or confidential data each AI feature uses, and why.
- Choose providers and hosting that match your privacy obligations.
- Make sure AI respects the same access rights as your people.
Transparency
- Tell customers when they are talking to an AI assistant.
- Show sources for factual answers where possible.
- Make it easy to reach a person.
Human oversight
- Name an owner for every AI system in production.
- Require human approval for decisions with significant impact on people or money.
- Give staff a simple way to flag and correct AI mistakes.
Testing and monitoring
- Test on real examples, including edge cases and unfair outcomes, before launch.
- Monitor quality, cost and complaints after launch.
- Re-test whenever the model, prompts or data change.
Records
Keep a short register of AI systems in use: purpose, data, owner, risk level and last review date. It takes an hour to start, makes audits and customer questions easy to answer, and prepares you for regulations that increasingly expect this kind of documentation.