When agents act on their own, governance has to live in the data layer

The increasing autonomy of intelligent agents raises critical questions about how to prevent unauthorized actions. These systems require intelligent rules that adapt to the moment's context, as general rules may fail during urgent decision-making scenarios. For example, a rule against opening a car door could hinder an agent's response in emergencies. Governance must be executable at the moment of decision-making, as agents interact directly with data. Policies need to ensure that agents cannot access certain data classes at the precise moment they request it, making real-time enforcement essential for effective governance.