Microsoft on Monday introduced new security tools meant to protect companies from risks tied to AI agents. The company says the agents businesses use could act like insider threats.
The main addition is Guardian Agents, a feature that watches over the AI agents a company has deployed. It is part of Project Perception, a security product Microsoft launched in June.
How Project Perception Works
Project Perception was Microsoft’s response to powerful hacking models. These include Anthropic’s Mythos and Fable and OpenAI’s vulnerability scanning system, Daybreak.
The product first launched with three classes of agents. A “red team” acted like hackers, a “blue team” detected those attacks and assessed weaknesses, and a “green team” fixed them.
The original design assumed AI hackers would come from outside a company. The latest update shifts focus to agents already working inside businesses.
Microsoft Leaders Warn Against Trusting AI Models
Earlier this month, Microsoft CEO Satya Nadella wrote on his private blog that companies should stop trusting AI model providers. He said this applies even when those providers promise safety and security.
He called for separate systems, not controlled by AI labs, to watch and control the agents those labs’ products launch.
“We simply can’t outsource responsibility for what intelligence does on our behalf,” Nadella wrote. “We need to separate the supply of intelligence from the authority over it.”
Hayete Gallot, who leads Microsoft’s security business, was more direct when announcing Guardian Agents. “Unfortunately, frontier models, regardless of their origin, cannot be trusted,” she wrote.
Gallot said this is not because the models are malicious or easy to hack. She said they are unpredictable and trained to complete tasks, not to follow “organizational norms or judgment.”
She also noted that these agents often have access to data and tools that could cause serious harm.
The warning comes from a company with close ties to AI labs. Microsoft has invested in both OpenAI and Anthropic, OpenAI is one of its largest cloud customers, and Microsoft runs its own AI lab.
Microsoft is offering its own agent security products as a cloud provider, separate from the companies that build the agents.
Beyond Guardian Agents, Microsoft added an agent detection feature to Microsoft Defender. It can find up to 60 types of agents that employees bring to work, including Claude Code, Codex, and OpenClaw.
The company also added features that limit agents’ access to data and systems. These tools adjust permissions based on the task and can require human approval when an agent appears to make risky decisions.
Microsoft was an inaugural member of Nvidia’s new Open Agent Safety Platform. The platform has support from more than 100 other tech companies.
It isolates agents in secure environments and can monitor and stop their behavior through hardware, working like a kill switch.
This follows Microsoft’s announcement of its new Surface Laptop Ultra PCs. The machines run on Nvidia chips and come with agent security and controls built in.
