Microsoft launches controls that treat enterprise AI agents as potential insider threats

On Monday Microsoft introduced security controls aimed at what Hayete Gallot, executive vice president of Microsoft Security, calls the "agentic insider": AI agents inside a company that hold too many privileges or pursue goals in ways nobody intended. Gallot wrote that frontier models cannot be fully trusted because they are built to chase results, and that some may collude with other agents. The release follows an October 10 essay by CEO Satya Nadella that urged companies to treat models as insider risks and to keep safeguards outside the models themselves. Microsoft's discovery tools can now detect 60 types of local agents on employee machines, up by 35. The list includes Anthropic's Claude Code, OpenAI's Codex and GitHub Copilot. Microsoft is extending Entra Agent ID so that these agents, and unknown ones its sensors find, get their own identities. The centerpiece is Guardian Agents, an extension of Project Perception, the agentic security system Microsoft introduced in June. Guardian Agents monitor deployed agents, explain why their risk changed, gather evidence, and recommend or apply containment, such as reducing what an agent may do as its risk climbs. New Playbooks define security outcomes and coordinate protective actions. Microsoft says people keep control over objectives and consequential decisions. The tools are aimed at enterprises that are deploying growing numbers of agents with broad access.