2026-11-04 · Fonte：InfoWorld
Microsoft leans on hardware virtualization to give AI agents a controlled runtime, addressing the trust problem that stalled the first wave of agent software in the 1990s.
Trust has become the central engineering problem for developers building AI agents, particularly those running on edge systems that blend local and cloud AI. The challenge is ensuring that code executing on a user's PC never touches private data or calls APIs it was never meant to reach. Agents are inherently unpredictable: they orchestrate calls across local and remote systems, assembling and running workflows on the fly. That unpredictability is precisely what makes them useful — and precisely what makes them hard to control.
The dilemma is not new. In the 1990s, the first generation of agents were tools designed to manage and run long workflows on behalf of users. They never achieved wide adoption, and the reasons are instructive: the security risks of delivering arbitrary code to PCs and servers proved too great, and the runtimes those agents required were too complex to build and share.
What has changed over the intervening decades is the arrival of hardware support for virtualization, a capability Microsoft has woven through much of its modern platform. Hardware virtualization underpins secure containers, Windows' own internal isolation model for critical operating system components, and cross-platform tooling such as the Windows Subsystem for Linux.
That foundation is now being extended to agent workloads. New tools such as the WSL container platform build on the virtualization model with SDKs that allow programs to launch and control the virtualization life cycle programmatically. It is this programmatic handle on hardware-isolated environments that Microsoft's Execution Containers bring to agent development, offering a way to run unpredictable agent code inside a boundary the host controls. For teams deploying agents on edge machines, the approach reframes the old trust question: rather than trying to predict what an agent will do, developers can guarantee where it is allowed to do it.
Tags: Microsoft · Execution Containers · Windows Subsystem for Linux · AI Agents · Sandboxing
萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado