Microsoft releases an open-source standard to control AI agents

In short: Microsoft has released an open-source standard called Agent Control Specification to help companies set clear rules for what AI agents are allowed to do.

What happened

Microsoft announced the Agent Control Specification, or ACS, a new open-source standard for controlling the behavior of AI agents. AI agents are AI tools that can take actions on your behalf, like using apps, calling other tools, or completing multi step tasks.

ACS lets developer, security, and compliance teams write “policy files” that spell out the rules an agent must follow. Think of it like a written rulebook that travels with the agent, even when the agent is used in different apps or systems.

Those rules can say what the agent is allowed to do, what it must not do, and when a person needs to approve an action. The rules can also require the agent to record evidence of what happened, so teams can review it later.

Microsoft says these checks can happen at multiple points while the agent works, such as before it takes in information, before it uses a tool, after a tool returns a result, and before it sends a final answer. Depending on the rule, ACS can allow an action, block it, or hide sensitive details (like blacking out a line in a document).

ACS is shipping as a software kit that can plug into several popular agent building tools, including LangChain, OpenAI’s Agents SDK, Anthropic’s Agents SDK, AutoGen, CrewAI, and Microsoft’s own Semantic Kernel.

Why it matters

As more businesses use AI agents for real work, mistakes can have real consequences, like sending the wrong data or taking an unintended action. A shared rulebook that is easier to reuse and audit could make these systems easier to supervise and safer to run.

Source: TechCrunch AI