- Where does AI governance start?
- With an inventory. Every framework, from NIST AI RMF to ISO 42001 to the EU AI Act, assumes you can enumerate the AI systems in use and say what each one does with what data. Organizations that skip straight to policy end up governing a list they cannot produce.
- Who authorizes an AI agent to act?
- Someone has to, and in most organizations no one has been named. An agent that can spend money, change configuration, or read customer data is exercising an entitlement, which means it needs an owner, an approval record, and a revocation path, exactly like a human account with the same reach.
- Do NIST AI RMF, ISO 42001, and the EU AI Act require different things?
- The three overlap more than the acronyms suggest. All three assume an enumerated inventory of AI systems, a documented owner per system, and a record of what data each one touches. The EU AI Act adds obligations that vary by risk classification and carry statutory deadlines, which is what makes it the one with a date attached.