Microsoft Says AI Agents Can Act Like Insider Threats, Launches Guardian Agents
Microsoft on Monday unveiled new security tools, including Guardian Agents, which monitors the AI agents a company uses. The firm argues that frontier AI models cannot be trusted because their behavior is unpredictable.
Technology company Microsoft on Monday, October 19, introduced new security features meant to protect businesses from artificial intelligence (AI) agents, which are programs that carry out tasks autonomously on a company's behalf. The main addition is Guardian Agents, a tool that monitors the AI agents an organization uses. Microsoft argues that such agents can behave much like an untrustworthy employee, a risk known as an insider threat.
What Guardian Agents is
The tool is part of Project Perception, which Microsoft introduced in June as its 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 imitated attackers, a "blue team" that detected them and assessed the vulnerabilities found, and a "green team" that fixed them. The design assumed attackers would come from outside the company.
Distrust of model providers
Microsoft CEO Satya Nadella warned in his private blog earlier this month that companies should stop relying on AI model providers' assurances of safety. He wrote that responsibility for what AI does cannot be outsourced, and that the supply of intelligence should be separated from authority over it. Hayete Gallot, who heads 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: they are trained to achieve tasks, not to follow organizational norms or judgment. They are also often given access to data and tools that could cause serious harm.
Other additions
Microsoft's Defender product now has a feature that can detect up to 60 types of agents employees bring into the workplace, including Claude Code, Codex and OpenClaw. The company also added controls that limit agents' access to data and systems, adjust their permissions dynamically based on 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, which has the support of more than 100 other tech companies. The platform isolates agents in secure environments and can monitor and stop their behavior through hardware, functioning like a kill switch. Microsoft has invested in both OpenAI and Anthropic, so the stance cuts against some of its own business ties, though it also positions the company to sell agent security products itself.
