Microsoft launches open-source standard to give developers granular control over AI agents

Microsoft has introduced a new open-source specification designed to help developers enforce consistent, auditable controls over AI agent behavior. Called the Agent Control Specification (ACS), the standard aims to address a growing challenge for enterprises deploying AI agents across diverse applications, workflows, and environments: ensuring agents act reliably and safely without fragmented or ad hoc governance.

What is the Agent Control Specification?

ACS is a policy framework that lets developer, compliance, and security teams define rules governing what an AI agent may do, what it must avoid, when human approval is required, and what evidence should be logged for review. These policies are written as single files that can be bundled with an agent, allowing governance to travel with the agent across different frameworks and deployment environments.

The specification checks agent behavior at multiple interception points during task execution: before receiving input, before calling a tool, after a tool returns a result, and before sending the final response to the user. At each point, a policy may allow an action, block it, redact sensitive information, or escalate to a human for approval.

Why this matters for enterprise AI

As organizations race to integrate AI agents into production systems, ensuring reliable behavior has become a critical concern. Current approaches often involve mixing instructions in system prompts, adding custom checks in application code, or using classifiers to catch problematic inputs and outputs. While these methods work, they frequently result in fragmented controls that are difficult to audit and hard to reuse across different frameworks, interfaces, and systems.

ACS aims to consolidate these controls into a common governance layer. Microsoft says the specification is designed to work with multiple frameworks, including LangChain, the OpenAI Agents SDK, the Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, and MCP tools. The SDK is shipping now with plugins for these environments.

Key capabilities and technical details

Developers using ACS can insert classifiers for inputs and outputs to categorize information, predict outcomes, or determine how an agent should respond. They can also add LLMs with prompts to act as a policy judge, along with logic for checking tool calls, tool selection, input accuracy, output usage, and final responses. Because policies are written as single files, they can be bundled directly with agents, ensuring a security policy follows the agent across different frameworks and environments.

The approach addresses real-world concerns about AI workflows going wrong due to tool misuse or unintended actions that cascade into failures. By providing a standardized way to define and enforce guardrails, ACS gives enterprises a more predictable path to deploying AI agents at scale.

Industry context and implications

The launch of ACS comes at a time when the AI industry is actively debating best practices for agent safety and governance. Several high-profile incidents involving AI agents taking unintended actions have highlighted the need for more robust control mechanisms. Microsoft’s move to create an open standard, rather than a proprietary solution, signals an intent to drive industry-wide adoption and interoperability.

For developers, the specification offers a way to reduce the complexity of managing agent behavior across multiple platforms. For compliance and security teams, it provides a consistent audit trail and policy enforcement mechanism that can be reviewed and updated independently of application code.

Conclusion

Microsoft’s Agent Control Specification represents a significant step toward standardized AI agent governance. By providing a portable, auditable policy framework that works across popular agent development frameworks, ACS addresses a practical need for enterprises deploying AI agents in production. The open-source nature of the specification may encourage broader adoption and community contributions, potentially shaping how the industry approaches agent safety and control in the years ahead.

FAQs

Q1: What is the Agent Control Specification (ACS)?
A: ACS is an open-source standard from Microsoft that lets developers define granular policies for AI agent behavior. It checks agent actions at multiple points during task execution and can allow, block, redact, or escalate actions based on predefined rules.

Q2: Which frameworks does ACS support?
A: ACS ships with plugins for LangChain, OpenAI Agents SDK, Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, and MCP tools, with more expected.

Q3: How does ACS differ from existing agent control methods?
A: Unlike ad hoc approaches that mix instructions in system prompts or application code, ACS provides a single, portable policy file that can be bundled with an agent and enforced consistently across different environments, making governance auditable and reusable.