微软推出Microsoft Execution Containers 1.0.0版本

微软正式发布Microsoft Execution Containers（MXC）1.0.0版本，这是一个针对AI代理的策略驱动执行容器1。该平台通过定义和强制执行资源访问边界，限制AI代理能够访问和执行的操作，为开发者和IT管理员提供统一的配置架构来管理代理安全性1。

MXC提供三种操作模式：强制执行模式、学习模式和宽容模式1。该技术支持多个操作系统环境，包括Windows、macOS和Linux，其中Windows 365现已支持MXC1。GitHub Copilot、OpenAI Codex、Replit、LM Studio和Unsloth AI等多个代理和框架已完成集成1。微软计划进一步扩展支持范围，包括Anthropic Claude Code、Box、Egnyte、Heidi Health、Perplexity和Raycast等代理1。此外，Windows将与Microsoft Entra配合使用以区分代理活动与用户活动，并通过Microsoft Intune实现策略管理1。

Microsoft has released Microsoft Execution Containers (MXC) version 1.0.0, a policy-driven execution platform designed to enhance the security of AI agents 1. The technology enables developers and IT administrators to define and enforce resource access boundaries, controlling which operations AI agents can perform and what data they can access through a unified configuration architecture 1.

MXC operates in three distinct modes to accommodate different security requirements: Enforcement mode for strict policy adherence, Learning mode for behavioral analysis, and Permissive mode for flexible operations 1. The platform supports multiple backends and operating systems including Windows, macOS, and Linux, with Windows 365 now offering native MXC support 1. Several prominent AI agents and frameworks have already integrated the technology, including GitHub Copilot, OpenAI Codex, Replit, LM Studio, and Unsloth AI 1. Additional integrations are planned with Anthropic Claude Code, Box, Egnyte, Heidi Health, Perplexity, and Raycast 1.

To strengthen agent management capabilities, Windows will leverage Microsoft Entra to distinguish between agent activity and user activity, while Microsoft Intune will provide policy management across enterprise environments 1.