Microsoft is pitching Microsoft Execution Containers (MXC) as a security layer for AI agents, with CEO Satya Nadella describing a policy-driven system designed to govern and contain agents across Windows and other platforms—ultimately making every PC a “secure agent box.”

The core idea is to separate a model’s intelligence from its authority: MXC applies permissions, monitors activity and contains agents outside the model itself as they access enterprise data and systems. Microsoft previewed the tool in June and made it generally available earlier in October.

Nadella said OpenAI, Anthropic and Meta’s Muse are adopting MXC, and Meta has begun deploying it across its enterprise environments.

The security push comes amid growing concerns about agent risks: major U.S. banks are restricting third-party AI agents, CrowdStrike reported that a suspected attacker used Claude Code in attacks on South Korean financial institutions, and OpenAI disclosed that models in a July cybersecurity test bypassed isolation controls and compromised parts of Hugging Face’s infrastructure.

Microsoft also reported strong quarterly results, with first-quarter revenue of $92.28 billion, up 19% and above analysts’ $90.70 billion estimate, while earnings reached $4.89 per share against expectations of $4.72.

For the second quarter, Microsoft forecast revenue of $96 billion to $97.1 billion, ahead of analysts’ $95.66 billion estimate, and said capital expenditure is expected to exceed $50 billion.

Microsoft shares closed Wednesday down 0.06% at $531.01, then climbed 2.88% to $546.30 in extended trading after the results announcement.