Microsoft launches MXC, an OS-level sandbox for AI agents, with OpenAI and Nvidia already on board

The developer then instructed the agent to delete a specific file, and again, the sandbox prevented it from doing so. The developer explained that this level of control is essential for ensuring that AI agents can operate safely within an enterprise environment without posing a security risk.

The implications of MXC are far-reaching. By providing a secure execution environment for autonomous AI agents, Microsoft is enabling enterprises to harness the power of AI without compromising the security of their systems. With MXC, developers and IT administrators can define precisely what resources an agent can access, ensuring that sensitive data remains protected.

Moreover, MXC’s ability to scale from lightweight process isolation to full virtual machines means that it can accommodate a wide range of workloads, from simple coding assistants to complex AI models. This flexibility makes MXC a versatile tool for enterprises looking to deploy AI agents in a secure and controlled manner.

In conclusion, Microsoft’s introduction of Microsoft Execution Containers at Build 2026 represents a significant step forward in the field of AI security. By providing a policy-driven execution layer within the Windows operating system, Microsoft is addressing the critical question of what happens when an AI agent goes wrong. With MXC, enterprises can now deploy autonomous AI agents with confidence, knowing that their systems are protected by a robust security framework. Microsoft’s MXC, or Microsoft Execution Containers, is set to revolutionize the way AI agents operate on Windows devices. With the ability for users to mark specific files as read-only for the agent, restrict access to the browser and screen capture, control location data visibility, and have all permissions managed centrally by an enterprise IT department through Intune policies, MXC ensures that agents can operate securely within defined boundaries.

Pavan Davuluri, Microsoft’s Executive Vice President for Windows and Devices, highlighted during a pre-briefing that the security, containment, isolation, and user control features introduced by MXC are crucial for making AI agents commercially viable. These capabilities are not unique to OpenClaw, the platform on which MXC is built, but rather a pattern that can be applied to any agent running on a Windows device. The existing primitives in the Windows operating system around security, containment, and user control are what will make agents safe for both consumers and corporate deployments.

One of the most significant elements of the MXC announcement for corporate IT departments is the integration with Microsoft’s enterprise security stack through Agent 365. Arriving in preview in July, Agent 365 layers Entra identity service and Intune device management platform on top of MXC. This integration allows IT administrators to govern agent containment centrally while developers choose the level of isolation required for their workload. Microsoft Defender provides threat protection, Entra handles identity management, Intune enforces device-level policies, and Microsoft Purview extends data governance and compliance capabilities to agent activity. This centralized visibility and control over agent activity could enable enterprises to run AI agents on corporate machines securely.

Several key partners, including OpenAI, Nvidia, Manus, and Nous Research, are already building on MXC, showcasing the diverse use cases for the technology. OpenAI is exploring new patterns for AI agents to generate and execute code safely, while Nvidia is bringing its OpenShell framework to Windows built on MXC. Manus and Nous Research are integrating MXC to enable developers to define agent access boundaries and ensure safe operation in enterprise environments.

The partnership with OpenClaw, an open-source agent framework, serves as a proving ground for MXC’s capabilities. Microsoft developers have contributed to the OpenClaw Windows companion app, testing the platform’s robustness in containing agents securely. If OpenClaw can operate successfully within MXC’s boundaries, then the containment system is considered robust for any agent. The companion app showcases MXC’s enterprise controls, such as file permissions, network access, screen capture restrictions, and location data management, all manageable through Intune policies.

Overall, Microsoft’s MXC is poised to set a new standard for AI agent security and control on Windows devices, paving the way for safe and efficient agent deployment in both consumer and corporate environments. Microsoft’s latest innovation, the MXC, is set to revolutionize the way AI agents are deployed in enterprise environments. This groundbreaking technology comes at a time when the industry is focused on developing AI agents, which are considered the most significant new category of software since mobile applications. Every major tech company is in a race to build these agents, but the lack of security and governance infrastructure is a major concern.

One key advantage that Microsoft has over its competitors like Apple and Google is its strategic approach to building containment into the operating system itself. By integrating security measures into Windows, Microsoft ensures that the security guarantees hold regardless of which agent, model, or framework a developer chooses to use. This unique approach sets Microsoft apart from Apple’s walled-garden ecosystem and Google’s cloud-first model.

With MXC, Microsoft is not only providing a secure platform for AI agents to run but also ensuring that enterprises can easily adopt this technology without the need for a complete overhaul of their existing systems. This means that Windows devices managed through Intune and secured through Defender can seamlessly transition to being agent-ready through a simple software update.

The competitive landscape is shifting with OpenAI’s Codex, Nvidia’s OpenShell, and independent agent frameworks like Manus and Hermes building on MXC. Microsoft is positioning Windows as the platform where agents can be trusted to run, catering to enterprises with diverse toolchains and multiple AI providers.

While MXC is currently available in early preview for developers to test containment policies, the real challenge will come when enterprises start deploying agents at scale on production networks. Writing effective agent policies for complex enterprise environments will be a new discipline that IT departments will need to develop. The success of MXC will ultimately depend on the ability to create and enforce the right policies for the right agents in the right contexts.

Microsoft’s announcement marks a significant milestone in the industry, as it provides a comprehensive solution for containing, identifying, and governing autonomous AI software on devices where most of the world’s work is conducted. The focus is now on teaching the operating system to monitor and manage AI agents effectively, a task that presents a new set of challenges for organizations. Despite the complexities involved, Microsoft’s MXC has the potential to reshape the way AI agents are deployed and managed in enterprise environments.