Microsoft casts AI agents as insider threats in new enterprise security push

Microsoft reframes enterprise AI risk

Microsoft on Monday introduced security features aimed at protecting enterprises from AI agents, as the company argues that frontier models cannot be trusted to act within organizational norms. The launch expands Project Perception, introduced in June, which originally organized agents into a red team that attacked, a blue team that detected and assessed vulnerabilities, and a green team that fixed them.

Microsoft security business chief Hayete Gallot said frontier models, regardless of origin, are not necessarily malicious or easily hacked, but are unpredictable because they are trained to achieve tasks rather than follow judgment. She said enterprises often give agents access to sensitive data and tools that could do serious harm.

Guardian Agents and Defender detection

The new Guardian Agents feature is intended to monitor the AI agents a company uses. Microsoft also added agent detection to Microsoft Defender that can find up to 60 types of agents employees bring into the workplace, including Claude Code, Codex and OpenClaw.

Microsoft said it is adding controls that limit agent access to data and systems, adjust permissions dynamically by task, and require human approval when an agent appears to be making risky decisions.

Separating intelligence from authority

Microsoft CEO Satya Nadella has warned that the AI enemy may be the very agents enterprises deploy and pay for, comparing them to a rogue human employee or insider threat. In a private blog post earlier this month, he said model provider assurances do not relieve enterprises of responsibility for what intelligence does on their behalf, adding that the supply of intelligence must be separated from authority over it.

Microsoft was an inaugural member of Nvidia’s Open Agent Safety Platform, backed by more than 100 other tech companies. The platform isolates agents in secure environments and can monitor and stop behavior through hardware. The move follows Microsoft’s announcement of Surface Laptop Ultra PCs built for agents, running on Nvidia chips with agent security and controls included.

Sources: TechCrunch