Microsoft rolls out MXC agent security tool

Microsoft has introduced Microsoft Execution Containers, or MXC, as a policy-driven execution layer designed to govern and contain AI agents across Windows and other platforms. On a first-quarter earnings call, CEO Satya Nadella described MXC as central to the company’s strategy of turning every PC into “the secure agent box.” Microsoft previewed MXC in June and made it generally available earlier in October, moving it from test phase to broad deployment. The launch positions the tool as infrastructure for managing how autonomous and semi-autonomous agents operate on user and enterprise systems.

Nadella explained that MXC enforces how agents run by applying policies around execution, with the goal of controlling what agents can do on devices and in connected environments. By embedding this layer into Windows and other platforms, Microsoft aims to standardize how agents are governed irrespective of the underlying model or application. The secure agent box concept emphasizes that agent activity should be tightly contained on endpoints rather than loosely integrated into existing workflows.

As AI agents gain greater access to enterprise data and systems, Nadella said security becomes “even more critical.” Microsoft’s approach with MXC is to “separate intelligence from authority,” keeping the model’s reasoning distinct from the permissions and controls that govern what actions it can perform. Under this model, permissions, monitoring and containment are enforced outside the AI model itself, providing an independent layer of oversight.

This separation is intended to limit the operational scope of agents while still allowing them to leverage advanced intelligence. By enforcing containment externally, MXC seeks to reduce the risk that an agent’s outputs translate into unintended or unauthorized actions within corporate environments. The emphasis on monitoring also reflects the need to observe agent behavior in real time as organizations explore more automated workflows.

OpenAI, Anthropic PBC and Meta Platforms’ Muse are adopting Microsoft’s MXC tool for their agent deployments. Nadella said Meta has started deploying MXC across its enterprise environments, indicating early large-scale use of the containment layer. These adopters connect MXC to some of the most prominent AI models and applications currently in use, extending its reach beyond Microsoft’s own services.

The engagement from multiple AI developers suggests that MXC is being positioned as a shared security and governance layer for agent-based systems. By integrating MXC into their stacks, these organizations can rely on a common mechanism to constrain agents while they interact with sensitive data and infrastructure. This aligns with Microsoft’s vision of making PCs and enterprise devices consistent, secure environments for agent execution.

The MXC rollout occurs against a backdrop of strong financial performance for Microsoft. In the first quarter, revenue rose 19% to $92.28 billion, above an analyst estimate of $90.70 billion. The company previewed MXC in June and brought it to general availability earlier this month, adding a new AI-focused security and governance product to its portfolio during the same period.

Positioning MXC as generally available indicates that Microsoft considers the technology ready for broad enterprise adoption rather than limited pilot use. Combined with rising revenue, the introduction of MXC underscores the company’s ongoing investment in AI infrastructure that addresses operational and security requirements for advanced agents.