Microsoft says AI agents should be treated as insider threats

Microsoft has introduced security features designed to monitor and control AI agents used by enterprises, warning that even advanced models can behave unpredictably.

The new Guardian Agents feature is part of Project Perception, which Microsoft introduced in June. The project initially included three types of agents: a red team to simulate attacks, a blue team to detect them and assess vulnerabilities, and a green team to fix those vulnerabilities. Guardian Agents are intended to monitor agents a company uses.

Microsoft security business chief Hayete Gallot said frontier models “cannot be trusted” because they are unpredictable and trained to complete tasks rather than follow organizational norms or exercise judgment. Microsoft CEO Satya Nadella has also argued that companies should use monitoring systems separate from the AI providers that supply their agents, especially when agents can access sensitive data and take critical actions.

Microsoft has added an agent detection feature to Microsoft Defender that can identify up to 60 types of agents employees bring into the workplace, including Claude Code, Codex, and OpenClaw. Other features can limit agents’ access to data and systems, adjust permissions based on tasks, and require human approval for decisions deemed risky.

Microsoft also joined Nvidia’s Open Agent Safety Platform, backed by more than 100 other technology companies. The platform is designed to isolate agents in secure environments and monitor or stop their behavior through hardware. Microsoft’s announcement follows its introduction of Surface Laptop Ultra PCs built for agents, with Nvidia chips and agent security controls.