BRK390 Breakout Wed 18 Nov 11:30 PST
MXC: Deploying Agents You Can Trust
Demitrius Nelon, Stuart Schaefer, David Wiesen
Summary
From the recording
Microsoft presented MXC (Microsoft execution containers), an open source, cross-platform policy layer in which the operating system, rather than the agent, enforces agent containment. Backends span process, Windows session, micro VM and cloud. A demo ran the same micro VM locally and in Azure Container Instances. OpenAI said about half of its Windows users now get sandboxing through MXC, replacing around 50,000 lines of bespoke code. Policy language is not yet uniform across containers, and audit and learning mode tools are being added. Speakers said they are early in the journey; no GA or preview status, dates or pricing were stated.
Session description
Microsoft's catalogue description
One sandbox doesn't fit every AI agent. Explore Windows isolation options through demos of Microsoft Execution Containers. Learn how to choose the right execution environment based on workload, security, and policy.
Demo timestamps
Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.
- Skip to 0:19:07
Live demos of session isolation, microVMs and Azure-based agent execution
MXC integrations with OpenCode and OpenClaw show different containment models. An agent under a separate Windows account analyzes financial data in isolation from the user's profile. The same microVM technology runs in Azure Container Instances.