Microsoft launches Guardian Agents to monitor AI agents amid insider-threat warnings

Microsoft introduced Guardian Agents, a new security feature designed to monitor AI agents used by enterprises, positioning them as potential insider threats. The tool is part of Project Perception, which Microsoft launched in June to address security risks posed by AI systems, including those from Anthropic and OpenAI.

Project Perception originally included three classes of agents: a red team simulating hackers, a blue team detecting vulnerabilities, and a green team fixing them. Microsoft now emphasizes the need to monitor agents deployed by enterprises, reflecting a shift from external threats to internal risks.

Microsoft CEO Satya Nadella argued in a private blog post that enterprises cannot outsource responsibility for AI-driven actions, stating that model providers’ assurances do not relieve organizations of accountability. He called for separating the supply of intelligence from the authority over it.

Hayete Gallot, Microsoft’s security business chief, stated that frontier models, regardless of origin, cannot be trusted due to their unpredictability. She noted that these models are trained to achieve tasks, not to adhere to organizational norms or judgment, and often have broad access to sensitive data and tools.

Guardian Agents is complemented by features in Microsoft Defender that detect up to 60 types of agents, including Claude Code, Codex, and OpenClaw. The toolset also includes capabilities to limit agent access to data and systems, dynamically adjust permissions, and require human approval for high-risk decisions.

Microsoft’s stance contrasts with its close ties to OpenAI and Anthropic, both of which it has invested in and collaborates with as a major cloud customer. Despite these relationships, Microsoft is advocating for independent monitoring systems to mitigate risks associated with third-party agents.

The company’s approach aligns with broader industry trends, such as Nvidia’s Open Agent Safety Platform, which includes over 100 supporting companies. The platform isolates agents in secure environments and provides hardware-level monitoring and control mechanisms, similar to a kill switch.

Microsoft also announced Surface Laptop Ultra PCs designed for agents, built on Nvidia chips and incorporating agent security and control features.