Microsoft releases open source standard for AI agent control

Microsoft has released an open source standard called Agent Control Specification (ACS) to help developers control the behavior of AI agents—which are software programs capable of performing tasks autonomously. As enterprises deploy these agents across different environments, applications, and products, ensuring they perform as intended has become a primary challenge. The new specification allows developer, compliance, and security teams to define specific policies for these agents to follow. These rules can define what the agent is allowed to do, what it must not do, when a human should approve an action, and what evidence should be logged for later review.

The release of ACS comes as developers are currently improvising methods to control what their agents see and do. This occurs amid conversations focusing on AI workflows going wrong due to tool misuse, or unintended actions that result in cascading failures. Currently, teams often rely on fragmented controls—such as system prompts, custom application code, or input-output classifiers—which are difficult to audit and reuse across different systems.

According to Microsoft, ACS can be used to check agent guardrails at multiple interception points in a workflow. These policy checks can occur before an agent receives input, before it calls a tool, after a tool returns a result, or before a final response is sent to the user. A policy may allow an action, block it, redact sensitive information, or ask a person to approve it. Developers can also insert classifiers for inputs and outputs to categorize information, predict outcomes, or determine how an agent should respond. Additionally, they can add large language models with prompts to act as a judge for policies, alongside logic for checking tool calls, tool selection, input accuracy, output usage, and responses.

Because these policies can be written as single files, they can be bundled with agents. This allows a security policy to follow an agent across different frameworks and environments.

ACS is shipping as a Software Development Kit (SDK)—which is a collection of software development tools—with plug-ins for various frameworks. The SDK partners for ACS include:

• LangChain
• OpenAI
• Anthropic
• AutoGen
• CrewAI
• Semantic Kernel
• Microsoft.Extensions.AI
• MCP (Model Context Protocol)

Why it matters

Enterprises currently struggle to audit and reuse AI agent controls across fragmented environments; Microsoft’s ACS aims to solve this by providing a unified, reusable governance layer for agent behavior.