Microsoft: AI agents can become insider threats for companies
Microsoft on Monday introduced new security tools, including Guardian Agents, which monitor the AI agents a company uses. The company argues that frontier AI models cannot be trusted because their behavior is unpredictable.
US technology company Microsoft announced new security features on Monday, October 19, aimed at protecting businesses from risks posed by artificial intelligence (AI) agents. AI agents are programs that carry out tasks autonomously on a company's behalf. Microsoft argues they should be monitored by systems that are not owned by the AI developers themselves.
Guardian Agents and Project Perception
The main addition is Guardian Agents, which oversees the AI agents a company deploys. It is part of Project Perception, which Microsoft introduced in June. That product was the company's answer to powerful hacking models such as Anthropic's Mythos and Fable, and OpenAI's vulnerability scanning system Daybreak.
Project Perception originally included three groups of agents: a "red team" that attacked like hackers, a "blue team" that detected them and assessed the vulnerabilities found, and a "green team" that fixed them. The product assumed attackers would come from outside the company.
Trust in models
Microsoft CEO Satya Nadella wrote in his private blog earlier this month that enterprises should stop relying on the safety assurances of AI model providers. In his view, companies need separate systems, not controlled by the AI labs, to watch their agents. He said the supply of intelligence must be separated from authority over it.
Hayete Gallot, who leads Microsoft's security business, said frontier models cannot be trusted regardless of their origin. She said this is not because they are malicious or easily hacked, but because they are unpredictable and trained to complete tasks rather than follow organizational norms or judgment. They are also often given access to data and tools.
Other additions
Microsoft added a feature to its Defender product that can detect up to 60 types of agents employees bring into the workplace, including Claude Code, Codex and OpenClaw. It also added features that limit agents' access to data and systems, adjust permissions dynamically to the task, and require human approval when an agent appears to be making risky decisions.
Microsoft was also an inaugural member of Nvidia's new Open Agent Safety Platform, supported by more than 100 other tech companies. The platform isolates agents in secure environments and can stop their behavior through hardware.
