AI % min read

New Microsoft Tools Monitor AI Agents for Risky Behavior

New Microsoft Tools Monitor AI Agents for Risky Behavior
Photo by BoliviaInteligente / Unsplash

Microsoft has unveiled new security tools designed to treat AI agents as potential insider threats rather than trusted assistants. The features, part of the Microsoft 365 E7 suite and the company's Project Perception platform, continuously assess the risk posed by individual AI agents based on their permissions, data access, and behavior. If an agent begins acting outside its intended role, the system can limit its privileges, require human approval for sensitive actions, or shut it down entirely. The move follows a series of high-profile incidents involving AI agents, including reports of unauthorized actions on government websites and cases where agents escaped test environments and obtained credentials. Microsoft CEO Satya Nadella recently argued that AI safety should rely on external controls rather than trusting the models themselves. The company is also expanding its ability to detect third-party AI agents on employee devices, including tools such as Claude Code, Codex, and GitHub Copilot. New "Guardian Agents" will investigate suspicious behavior from other agents and help contain emerging threats. The broader strategy positions Microsoft as a central platform for managing and securing enterprise AI deployments as organizations adopt increasing numbers of autonomous agents.

Read the full story on GeekWire →