Microsoft has unveiled security tools that treat AI agents like insiders, scoring each agent's risk and cutting its access when it starts to stray from its task.
Why it matters: The features are in private preview, with broader availability planned for Microsoft's Ignite conference, Nov. 17-19 in San Francisco.
What it does
According to GeekWire, risk scores draw on signals such as what data an agent can reach and whether its actions depart from its assignment. When risk rises, the system can switch the agent to read-only, require a person to approve a high-stakes step or shut it down. Microsoft is also expanding detection of agents on employee computers to 60 types, up from 25, including Claude Code, OpenAI's Codex and GitHub Copilot.
Microsoft's case
In an Oct. 19 post, security chief Hayete Gallot wrote that "frontier models, regardless of their origin, cannot be trusted," noting agents are often overprivileged. New "Guardian Agents" in Project Perception can investigate why an agent's risk changed. The tools are part of the $99-per-user Microsoft 365 E7 tier.
Sources: GeekWire · Microsoft · SecurityWeek