Microsoft has introduced new security measures for AI agents, warning that these automated tools could pose as much of a threat as rogue employees. The Guardian Agents feature, part of Project Perception, aims to monitor AI agents used by enterprises, highlighting the unpredictability of these models, which are often granted access to sensitive data.

Microsoft's CEO, Satya Nadella, has emphasized the need for enterprises to implement separate, independent systems to oversee AI agents, citing the unpredictability of these frontier models. He warns that even providers with safety assurances cannot be fully trusted, advocating for a separation between the supply and authority of intelligence.

The move is both surprising and strategic for Microsoft, which has invested in and competes with some of the very companies now facing increased scrutiny. By launching these security features, Microsoft is positioning itself to provide a solution, encouraging enterprises to buy their agent security products rather than relying on competitors.

Other tech giants, including Nvidia, are aligning with this approach, through platforms like the Open Agent Safety Platform, which isolates agents in secure environments and monitors their behavior. This industry-wide shift reflects a growing concern about the potential risks of AI, even among those who benefit from its development.