Microsoft unveiled a set of new security features on Monday aimed at protecting enterprises from the risks posed by AI agents, TechCrunch reported. The headline addition is Guardian Agents, a tool that monitors the AI agents a company deploys and runs as part of Project Perception, a product Microsoft introduced in June.
The launch doubles down on a warning Microsoft leadership has been repeating for months: the biggest AI danger may not come from outside hackers, but from the very agents companies pay for and hand access to. Microsoft security chief Hayete Gallot put it bluntly, writing that frontier models, whatever their origin, cannot be trusted.
Guardian Agents and the Project Perception expansion
Project Perception was originally designed around the idea that AI hackers would strike from outside. It launched with three classes of agents: a red team that attacks like a hacker, a blue team that detects and assesses the vulnerabilities found, and a green team that fixes them. Guardian Agents now shifts that logic inward, watching over the agents enterprises themselves are running.
Microsoft is pitching the tool as separate from the AI labs whose models it monitors, so that the software policing agent behavior does not come from the same suppliers that built the agents. According to TechCrunch, the company frames this as a matter of independence, much like auditors staying separate from the organizations they audit.
Microsoft Defender now detects up to 60 types of AI agents
Alongside Guardian Agents, Microsoft added an agent detection feature to Microsoft Defender that can spot up to 60 kinds of agents employees bring into the workplace, including Claude Code, Codex, and OpenClaw. The company also introduced controls that limit what agents can reach, adjust their permissions based on the task at hand, and require human approval when an agent appears to be making risky decisions.
