AI Agents as Insider Threats: Inside Microsoft's Bet on a Unified Security Stack

Microsoft is releasing a new feature within Project Perception, called Guardian Agents, to address the growing risk of AI agents behaving unpredictably and potentially causing harm within organizations. While Microsoft emphasizes the importance of a layered approach to security, including integrating signals from third-party products, some analysts remain skeptical, arguing that relying heavily on a single vendor’s ecosystem can limit effectiveness and that true agent security requires a broader, open-standard approach. The release is currently limited to Microsoft and select customers, and the feature is part of the more expensive Microsoft 365 E7 suite.