As a lawyer watching this space, I'd double-click on this: "we must understand what they’re doing, why they’re doing it, what has been affected and, critically, how to stop and/or recover it."

Explainability and auditability have to be solved at this foundational, stateful layer. If engineering doesn't get this right natively, the market is going to be flooded with rubbish, bolted-on regtech 'solutions' that try to achieve compliance by surveilling the user and/or simply guessing at the agent's logic.

Great points and so relevant to agents in production!