Microsoft launches Execution Containers to secure AI agents

Microsoft has released Microsoft Execution Containers (MXC), a security toolkit designed to manage and restrict the access of AI agents within operating systems. Unveiled at the Build 2026 conference, MXC provides a policy-driven Software Development Kit (SDK) that allows developers and IT departments to establish boundaries for file and network access.

The technology utilizes a “composable sandbox” approach, offering different levels of isolation including process isolation, session isolation, and MicroVMs. These boundaries are enforced by the operating system kernel in real time to prevent unauthorized data access and implement the principle of least privilege. MXC supports Windows, macOS, Linux, and the Windows Subsystem for Linux (WSL).

Early adopters of the technology include GitHub Copilot and the open-source framework OpenClaw. Other partners expected to utilize the containers to enable enterprise-level rule enforcement include NVIDIA, Anthropic, and OpenAI. Microsoft plans to integrate MXC with its existing security and management ecosystem, including Microsoft Entra and Microsoft Intune.