Microsoft is arguing that AI agents need security primitives built into the platform they run on, not just safeguards inside the model. InfoQ highlights the Microsoft Execution Containers SDK as a key part of that strategy.
The approach treats agents as active software that may need isolation, identity controls, and manageability from the operating system. That is a different security posture from simply letting an assistant run inside a normal user session.
As agent tools gain more autonomy, platform-level containment could become one of the default requirements for enterprise adoption.