Bottom line: Microsoft creates dedicated security frameworks for autonomous AI agents with the Execution Container and MDASH system to prevent uncontrolled access, data leaks, and code execution.
Microsoft presents new security tools for AI agents at its Build developer conference, including a proprietary runtime container called Microsoft Execution Container (MXC). These measures aim to control the risks associated with autonomous agents in development workflows.
As enterprises integrate AI agents into their software development workflows, Microsoft is introducing new security controls. The Microsoft Execution Container (MXC) is a dedicated sandboxing technology for autonomous AI agents. It enables developers to define via JSON configuration and TypeScript SDK which files, networks, resources, and credentials an agent may access, and enforces these boundaries at runtime. MXC runs on Windows, Linux, and macOS and offers various containment backends, from OS-native process sandboxes to fully virtual machines.
The threat is concrete: today, coding agents can access files they should not have permission to use, disclose secrets, make unauthorized network calls, and execute unexpected actions. Aleš Holeček, Chief Architect at Microsoft Security, names the core problems as “insecure code, opaque models, data exposure, and compliance violations.” The MXC integration into Agent 365, the new platform for managing AI agents, enables controls from Defender, Entra, Intune, and Purview to be brought directly into the agent environment.
In parallel, Microsoft is expanding its MDASH system (Security Multi-model Agentic Scanning Harness), originally introduced in May. The system deploys over 100 specialized AI agents to identify code vulnerabilities, assess exploitability, and reduce false-positive findings before they reach the security team. MDASH is now available in extended preview and integrates with Microsoft Defender. The system helped discover multiple Windows security vulnerabilities, including critical remote code execution flaws.
Microsoft also showcased open-source initiatives for AI agent governance. The new tools and platforms – including Windows 365 for Agents, a managed cloud environment for autonomous agents – are designed to give security teams a consistent view across the entire agent lifecycle. OpenClaw and NVIDIA’s OpenShell already integrate MXC into their agent execution.
Source: www.csoonline.com · Published June 3, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification through Lumi News Pipeline v1.2.9.