Most organizations have clear rules for who may access which system and who approves what. Agents built in Copilot Studio or Microsoft Foundry can act in the same systems, often with fewer of those rules applied to them.
An operating model for agents answers a short list of questions for every agent: who owns it, what it is allowed to read and change, how its actions are recorded, who handles its exceptions, and when it is reviewed or retired.
None of this requires waiting for new technology. Identity, data protection and monitoring capabilities already exist across Microsoft Entra, Purview and Defender. The work is deciding how they apply to agents, and assigning accountability.