Microsoft says AI agents are like insider threats as 'frontier models can't be trusted'
Microsoft wants enterprises to closely watch the agents employees use with systems not controlled by AI labs like OpenAI and Anthropic.
TL;DR
- Microsoft introduced Guardian Agents, a new feature to monitor AI agents used by companies, as part of Project Perception.
- Microsoft CEO Satya Nadella warns that AI agents, even those deployed by enterprises, could pose insider threats.
- Nadella advises against fully trusting AI model providers and recommends separate systems to monitor and control AI agents.
- Microsoft security chief Hayete Gallot stated that 'frontier models, regardless of their origin, cannot be trusted' due to their unpredictability.
- Microsoft's strategy involves offering its own agent security products, creating a potential conflict of interest given its investments in OpenAI and Anthropic.
- New features in Microsoft Defender can detect various employee-brought agents, and controls limit agents' data access and dynamically adjust permissions.
- Microsoft is also part of Nvidia's Open Agent Safety Platform to isolate and monitor AI agents.