Microsoft Introduces Open-Source Standard to Streamline AI Agent Governance

As artificial intelligence agents become increasingly sophisticated, organizations rushing to integrate them into various applications, workflows, and products are encountering a significant hurdle: maintaining precise control over agent behavior across diverse deployment environments. To address this issue, Microsoft has launched the Agent Control Specification (ACS), an open-source standard designed to provide developers with a more uniform and detailed method for regulating AI agent permissions.

At its core, ACS enables development, compliance, and security teams to establish custom policies that dictate agent conduct. These policies can specify permissible actions, prohibited activities, requirements for human approval, and protocols for logging evidence for future audits. To ensure agents remain within established boundaries, these policy files are evaluated at multiple "interception points" while the agent executes tasks.

The release of ACS responds to growing industry concerns regarding AI workflow failures, often triggered by tool misuse or unintended actions that lead to cascading errors. Currently, developers frequently resort to ad-hoc solutions, such as embedding instructions in system prompts, implementing custom checks within application code, or deploying classifiers to filter problematic inputs and outputs. While effective to some degree, these methods often result in fragmented control mechanisms that are difficult to audit and challenging to repurpose across different systems, interfaces, and frameworks. ACS seeks to consolidate these disparate controls into a unified governance layer.

According to Microsoft, the specification allows for continuous monitoring of agent compliance at critical stages of the workflow: prior to input reception, before tool invocation, following tool execution, and before the final response is delivered to the user. Policies can be configured to permit or block actions, redact sensitive data, or mandate human oversight. Additionally, the framework supports the integration of classifiers to categorize data, predict outcomes, and guide agent responses, as well as the inclusion of Large Language Models (LLMs) acting as policy "judges." It also facilitates logic for verifying tool calls, selection accuracy, input integrity, output usage, and response quality.

A key advantage of ACS is its portability. Because policies are stored in single files, they can be bundled directly with agents, ensuring that security protocols travel with the agent regardless of the framework or environment in which it operates. The ACS SDK is now available, featuring plug-ins for a wide range of tools and frameworks, including LangChain, the OpenAI Agents SDK, the Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, and MCP tools.

Source: TechCrunch Generated at: 2026-06-02 18:00:00 UTC