Microsoft says AI agents should be watched like insider threats

Microsoft on Monday introduced Guardian Agents, a feature that monitors the AI agents a company uses, as part of new security tools for businesses.

“Unfortunately, frontier models, regardless of their origin, cannot be trusted.”

— Hayete Gallot, Microsoft's security business chief

Why it matters

Chief executive Satya Nadella says companies need systems not controlled by the AI labs to watch their agents, because models are unpredictable and often have wide access to data and tools.

Other new controls

• Detection of up to 60 types of agents
• Limits on agents' access to data and systems
• Permissions adjusted to each task
• Human approval for risky decisions

Microsoft also added tools that detect up to 60 types of agents employees bring to work, limit their access to data and require human approval for risky decisions.