Microsoft launches Guardian Agents to monitor enterprise AI agents
Original title: Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’
Hayete Gallot says frontier models cannot be trusted because they are unpredictable, not malicious.
Microsoft introduced Guardian Agents, part of Project Perception, to monitor AI agents used inside enterprises. CEO Satya Nadella argues companies should not rely on model providers' safety assurances and should separate intelligence supply from authority over agents. Defender can detect up to 60 employee-brought agent types, including Claude Code, Codex and OpenClaw, and can limit permissions or require human approval. Microsoft also joined Nvidia's Open Agent Safety Platform.
Unconfirmed: Gallot's claim that frontier models cannot be trusted is Microsoft's position, not an independently measured finding.
- Project Perception originally shipped red, blue and green team agent classes.
- Nadella says model-provider assurances do not remove enterprise responsibility.
- Permissions can change with the task and require approval for risky decisions.
- The Open Agent Safety Platform has support from more than 100 companies.
- Surface Laptop Ultra PCs add agent security controls on Nvidia chips.
- Vendors
- Microsoft
From the original article
3 min read at techcrunch.com
Microsoft wants enterprises to closely watch the agents employees use with systems not controlled by AI labs like OpenAI and Anthropic.
By Julie Bort · techcrunch.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.