Microsoft Launched Seven AI Models and an Agent Sandbox on the Same Day

Frequently asked

What does the MXC isolated execution environment actually prevent agents from doing?

MXC creates a bounded environment for agent execution — limiting what data, tools, and systems an agent can access during a task. The practical effect is that an agent running inside MXC cannot reach outside its defined container without explicit configuration, which shifts the security perimeter from the application layer to the infrastructure layer. Microsoft controls what that perimeter looks like by default.

Why does Microsoft need proprietary models if it already distributes OpenAI models through Azure?

Distributing a partner's models creates dependency and margin pressure — Microsoft captures less value when the model is OpenAI's than when it is its own. Seven proprietary models give Microsoft negotiating leverage, a differentiated product line for enterprise contracts, and the ability to tune pricing independent of OpenAI's cost structure. The [open models Microsoft shipped earlier this week](/story/microsoft-shipped-two-open-models-week-neither-enterprise-34179489) and these proprietary ones serve different commercial purposes.

What should enterprise procurement teams know before committing to MXC?

Adopting MXC means accepting Microsoft's definition of agent security boundaries at the infrastructure level. Once MXC is embedded in enterprise contracts and workflows, renegotiating those defaults becomes a procurement and legal problem, not a technical one. Teams should document what the container permits and prohibits before signing, not after deployment reveals the gaps.

More on this wire