Microsoft warns AI agents can become “double agents” when permissions sprawl and security lags. Memory poisoning and deceptive inputs can steer tools off course, so it recommends Zero Trust controls, inventory, and continuous monitoring.
The post Micro…