Microsoft launches open‑source Agent Control Specification to tighten AI guardrails

Key Points

• Microsoft introduced the open‑source Agent Control Specification (ACS) for AI agents.
• ACS lets developers define policies that dictate allowed actions, prohibitions, human approvals, and logging requirements.
• Policy checks occur at four interception points: before input, before tool calls, after tool results, and before final output.
• The SDK includes plug‑ins for LangChain, OpenAI Agents SDK, Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel and more.
• ACS aims to replace fragmented, hard‑to‑audit controls with a reusable, auditable governance layer.
• Built‑in classifiers and LLM “judges” can evaluate inputs, outputs and policy compliance in real time.
• Policy files travel with agents across environments, ensuring consistent security posture.
• Early users report reduced risk of tool misuse and easier regulatory reporting.