Microsoft adds Guardian Agents for enterprise AI security

Microsoft says it is adding Guardian Agents and new risk-based controls to Project Perception, its security platform for organizations using AI agents. The company says the system can continuously assess an agent’s risk and adjust what it may access or do, including requiring human approval, isolating its runtime or stopping it. ^[1]

Microsoft says Guardian Agents can investigate why an agent’s risk changed and recommend or apply containment within policies set by people. The announcement describes planned capabilities, but the sources do not provide independent tests of how well the controls perform in practice. ^[1][2]