Microsoft has made Execution Containers (MXC) generally available on Windows 11, providing a policy-driven execution layer for untrusted code and AI agents. The technology allows developers to define which files and networks agents can access, with those policies enforced at runtime, marking a significant step in securing agentic AI workloads on the Windows platform and addressing growing concerns about the security implications of autonomous AI systems. The announcement came during a Windows and Surface news event in San Francisco, where Microsoft Chairman and CEO Satya Nadella and NVIDIA founder Jensen Huang showcased the future of hybrid intelligence on Windows.
Why Agents Need Managed Boundaries
AI agents can work across files, networks, and applications, introducing new security risks that traditional sandbox solutions weren’t designed to address. MXC provides containment (limiting what an agent can access and do), identity (distinguishing agent activity from a person), and manageability (giving organizations tools to govern access and monitor agent activity). Policies stay outside the agent workload’s control, preventing the agent or generated code from granting itself additional access, a concern that has grown more pressing following multiple publicly disclosed rogue agent hacks in 2026. Agents will also have the ability to explain when a task can’t be completed due to available permissions and request appropriate user or administrator action when supported.
Containment Options
MXC offers multiple containment backends suited to different workloads, giving developers flexibility to match security requirements with performance needs:
- Process container: Lightweight containment for responsive workloads, using AppContainer on Windows, Seatbelt on macOS, and Bubblewrap on Linux
- Session container: Long-running agents needing a desktop or stronger separation, running under a distinct Windows account with separate clipboard, UI, and input
- WSL container (WSLc): Linux-first agent toolchains that depend on the Linux development ecosystem and package management
- MicroVM: Higher-risk workloads with hardware-backed virtualized isolation and full Linux workload compatibility (currently experimental)
Ecosystem Support and Integration
Leading agents already support MXC, including OpenAI Codex, Microsoft GitHub Copilot, OpenClaw, Replit, LM Studio, NVIDIA OpenShell, and Unsloth AI. Anthropic Claude Code, Box, Egnyte, Heidi Health, Hermes Agent by Nous Research, Manus, Perplexity, Raycast, and Simular are among those planning to add support. Microsoft also announced that Meta’s Muse personal AI agent is coming soon as a native Windows app with MXC integration, bringing another leading AI experience to Windows and providing more choice in the agents people can use. Developers integrate with a unified JSON configuration schema and multi-language SDK, and MXC maps the requested controls to the selected back ends on Windows, macOS, or Linux, creating a simpler experience by separating workload requirements from platform-specific containment details.
Hybrid Intelligence on Windows
MXC is part of Microsoft’s broader “hybrid intelligence” strategy for Windows, where AI runs locally when it makes sense and reaches the cloud when needed. The Windows and Surface event on October 7 also showcased a new Windows Search experience, the Surface Laptop Ultra, and NVIDIA RTX Spark Dev Box. Yusuf Mehdi, Microsoft’s executive vice president, described Windows as “the platform for hybrid intelligence, where AI can run locally when it makes sense and reach the cloud when needed,” emphasizing that containment, identity, and manageability are built into the platform to help organizations control what agents can access. For individuals using AI on their personal PC, these safeguards are part of the agent experience itself, giving users a more secure way to work with these tools.
Microsoft also announced upcoming Intune policies for managing MXC process containers, giving IT administrators greater control over how Windows evaluates container creation requests made by agents. The company plans to integrate MXC with Microsoft Entra and Agent 365 to help distinguish agents from human users, a step toward Microsoft’s stated goal of licensing many agents similarly to the way they license people by providing them with their own Entra IDs, email addresses, and Teams accounts.
For more on the intersection of AI agents and consumer hardware, see our coverage of AI agents on consumer PCs in 2026.
Sources: Windows Developer Blog, October 7, 2026; Windows Experience Blog, October 7, 2026

Sources
- https://blogs.windows.com/windowsdeveloper/2026/10/07/microsoft-execution-containers-policy-driven-containment-for-ai-agents/
- https://blogs.windows.com/windowsexperience/2026/10/07/building-windows-for-hybrid-intelligence/





