Loading page…
RTX Spark laptops are due October 16, while the larger Windows workstation remains a preview. Microsoft’s containment layer is available now, but several management features are still coming.
Listen to this story
Microsoft released Execution Containers (MXC), which put developer-defined file, network and resource permissions outside an AI agent’s control. Its process containers support Windows 11, macOS and Linux, while desktop-session isolation is Windows-only. At the same event, NVIDIA detailed local-computing systems with up to 128GB of unified memory for RTX Spark and previewed a 748GB DGX Station for Windows, whose availability is unannounced. Agent containment is available now, but several enterprise attribution and management integrations are still forthcoming.
RTX Spark laptops are scheduled for October 16 availability, with compact desktops following in November.
NVIDIA says the DGX Station for Windows can run models up to trillion-parameter scale locally, but has not announced a release date.
MXC learning mode blocks and records ungranted access; permissive mode records policy violations but allows access to proceed.
NVIDIA and Microsoft paired more powerful local AI hardware with tighter agent permissions at their October 7, 2026, event in San Francisco. RTX Spark laptop preorders opened, a larger Windows workstation was previewed, and Microsoft released Microsoft Execution Containers—software designed to keep an agent’s access within boundaries set outside the agent itself.
RTX Spark laptops are scheduled to become available October 16, with compact desktops following in November. NVIDIA lists systems from Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte. The compact desktop design is intended for continuous operation, giving always-on agents a dedicated local machine.
DGX Station for Windows targets larger development workloads. NVIDIA says it can run models up to a trillion-parameter scale locally, but has not announced availability. Previously, DGX Station ran on Linux; the Windows version brings that computing environment onto the enterprise desktop, while retaining access to Linux tools through Windows Subsystem for Linux.
NVIDIA specifies up to 128GB of unified memory for its laptop and compact-desktop platform.
NVIDIA specifies 748GB of coherent memory for the previewed GB300 Grace Blackwell Ultra-based workstation.
Microsoft Execution Containers, or MXC, addresses what agents are allowed to do rather than how much computing power they have. Developers specify permitted files, network destinations and other resources. MXC selects a container—a restricted execution environment—to enforce those permissions. The policy stays outside the workload’s control, so generated code cannot grant itself more access.
Microsoft illustrates the distinction with a coding agent updating a website. It could receive permission to edit the repository and read production configuration, without permission to change that configuration. The boundary is designed to block an unauthorized change even if the agent decides it would help finish the task.
On Windows, MXC process containers can help developers discover which resources an agent needs. Learning mode blocks ungranted access and records attempts. Permissive mode records access that the policy would deny but lets it proceed, subject to other operating-system or organizational restrictions. Enforcement mode blocks unauthorized operations without producing the activity report.
Containment is available now; broader management is not all shipping yet. Microsoft says Entra-based attribution of agent activity, Agent 365 controls for local agents and Intune management policy for Windows 11 process containers are coming soon. Those additions are intended to let organizations govern agents separately from the employees using them.
Loading discussion...
Join the conversation
Explain when interrupted work is preferable to allowing unexpected access.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.