Microsoft Warns That AI Agents Pose Insider Threat, Launches Guardian Agents

Microsoft has rolled out new security tools aimed at protecting businesses from the risks posed by AI agents, while its executives increasingly warn that the technology cannot be trusted.

The centerpiece is Guardian Agents, a feature that monitors the AI agents a company deploys. It sits within Project Perception, a product Microsoft introduced in June with three agent classes: a "red team" that attacks like a hacker, a "blue team" that detects and assesses vulnerabilities, and a "green team" that fixes them.

That product assumed AI attackers would come from outside. But CEO Satya Nadella has warned that the real danger may be the agents enterprises pay for, likening them to a rogue employee, or an "insider threat." In a private blog post, he said model providers' safety assurances do not absolve companies of responsibility, and called for separate systems not controlled by the AI labs to watch over their agents. "We need to separate the supply of intelligence from the authority over it," he wrote.

Microsoft security chief Hayete Gallot put it bluntly: "frontier models, regardless of their origin, cannot be trusted." She said the problem is not malice or hackability but unpredictability, since the models are trained to complete tasks rather than follow organizational norms, even as they are given access to sensitive data and tools.

Microsoft's answer carries competing interests, given its investments in OpenAI and Anthropic and OpenAI's role as a major cloud customer. Still, the company is also adding agent detection to Microsoft Defender, covering up to 60 types of agents employees bring to work, along with features that limit agent access and require human approval for risky decisions. Microsoft was also an inaugural member of Nvidia's Open Agent Safety Platform, which isolates agents and can halt their behavior through hardware.

Source: TechCrunch