- Published on
Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’
- Authors
- Name
- DeepHorus
Microsoft on Monday unveiled new security features to protect enterprises from AI agent threats. The standout is Guardian Agents, which monitors the AI agents a company uses. It is part of Project Perception, launched in June.
Project Perception initially deployed red, blue, and green teams—agents that attack, detect, and fix vulnerabilities. It assumed hackers were external. Now CEO Satya Nadella warns the enemy may be enterprises’ own deployed agents, akin to insider threats.
Earlier this month, Nadella blogged that companies should stop trusting AI model providers’ safety assurances. He urged separate oversight systems not controlled by AI labs. “We need to separate the supply of intelligence from the authority over it,” he wrote.
Microsoft security chief Hayete Gallot said “frontier models, regardless of their origin, cannot be trusted.” Not because they are malicious, but because they are unpredictable, trained to achieve tasks rather than follow organizational norms. They often hold sensitive data and tools.
The solution is surprising and self-serving. Microsoft is close to OpenAI and Anthropic, having invested in both; OpenAI is a major cloud customer. Yet its logic is sound: monitoring software should be independent from agent suppliers, like auditors.
Microsoft added Defender detection for up to 60 agent types, including Claude Code, Codex, and OpenClaw. The feature scans for unsanctioned agents. New controls limit agents’ data access, adjust permissions dynamically, and require human approval for risky decisions.
The insider-threat view is spreading. Microsoft joined Nvidia’s new Open Agent Safety Platform as an inaugural member. As autonomous agents proliferate, demand for independent oversight and security controls will likely grow.
