Microsoft has helped companies build AI agents that can do specific kinds of work. Now it has a way to spot and stop risky agents.
