Microsoft announced in June 2026 an open source specification called Agent Control Specification (ACS), designed to give developers a more consistent and granular way to define what AI agents are and are not allowed to do.
The specification allows developer, compliance, and security teams to write policy files that govern agent behavior — defining permitted actions, prohibited actions, when a human must approve an action, and what should be logged for later review. These policies are checked at multiple points during an agent’s workflow: before it receives input, before it calls a tool, after a tool returns a result, and before a final response is sent to the user. A policy can allow an action, block it, redact sensitive information, or route it for human approval.
Microsoft is addressing a problem that has emerged as AI agents are deployed more widely across enterprise applications and workflows. Currently, developers control agent behavior through a mix of system prompt instructions, custom application code, and classifiers — approaches that can leave organizations with fragmented controls that are difficult to audit and reuse across different frameworks and systems. ACS aims to consolidate those controls into a common governance layer.
Because policies can be written as single files and bundled with agents, a security policy could follow an agent across different frameworks and environments. Developers can also add classifiers to categorize inputs and outputs, use large language models as policy “judges,” and apply logic to check tool calls, tool selection, and response accuracy.
ACS is shipping as an SDK with plug-ins for LangChain, the OpenAI Agents SDK, the Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI, MCP tools, and others. The move comes as the industry grapples with AI workflows failing due to tool misuse and unintended actions that can produce cascading failures.
Source: TechCrunch