Walk these steps in order. If you hit a “stop,” resolve it before continuing.
- Is the task allowed? If AI use would break a law, policy, regulation, member rule, contract, or license, stop. (Common examples: bypassing accessibility requirements, violating someone’s privacy, or producing prohibited content.)
- What’s the most sensitive data involved? Classify it using Section 4 and pick a tool approved for that level. If no approved tool fits, stop and talk to your IRM.
- Could the output drive a consequential decision? If it affects someone’s rights, grade, aid, job, benefits, care, or legal status, route it through Section 7 (human review and disclosure) before you proceed.
- Can I verify the result? If you cannot check the output for accuracy and bias, do not rely on it. Build in a verification step.
- Do I need to disclose? If a person is interacting with AI, or AI materially shaped a public communication or a decision about them, plan how you will tell them (Section 8).
- Is this new tool or vendor vetted? If you are introducing a new tool, especially one others will use or one that touches non-public data, run the procurement checklist in Section 10 first.