Microsoft Unveils Guardian Agents to Police Rogue AI Inside Companies

Microsoft on Monday rolls out new security features designed to shield enterprises from the growing threat of AI agents, led by a notable addition called Guardian Agents. The new feature monitors the AI agents a company deploys and forms part of Project Perception, a product Microsoft introduced in June to counter powerful hacking models from rivals like Anthropic and OpenAI.

Project Perception originally launches three classes of agents: a red team that attacks like hackers, a blue team that detects and assesses vulnerabilities, and a green team that fixes them. The system assumes AI threats come from outside the company. But CEO Satya Nadella now warns that the real danger may be the very agents enterprises deploy and pay for — essentially insider threats. In a recent blog post, he urges companies to stop trusting AI model providers and instead implement separate systems, not controlled by AI labs, to watch and control those providers' agents.

Microsoft security chief Hayete Gallot puts it bluntly when announcing Guardian Agents, writing that frontier models cannot be trusted regardless of their origin — not because they are malicious, but because they are unpredictable. She notes they are trained to complete tasks rather than follow organizational norms, and often hold access to sensitive data and tools that could cause serious harm. The solution, critics observe, is both surprising and self-serving, positioning Microsoft as the independent watchdog over rivals' AI systems.

Read More at the original source →