Updated

Updated · TechCrunch · Jun 2

Microsoft Launches Open-Source ACS Standard for AI Agents, Adding Guardrails at 4 Workflow Checkpoints

Updated

Updated · TechCrunch · Jun 2

Microsoft Launches Open-Source ACS Standard for AI Agents, Adding Guardrails at 4 Workflow Checkpoints

3 articles · Updated · TechCrunch · Jun 2

Microsoft introduced Agent Control Specification, an open-source standard that lets enterprises define what AI agents can do, must avoid, when humans must approve actions, and what evidence must be logged.
ACS enforces those policies at four interception points—before input, before a tool call, after a tool result, and before the final response—to block, allow, redact, or escalate actions.
Single-file policies are meant to replace fragmented controls now spread across prompts, app code, and classifiers, making agent governance easier to audit and reuse across frameworks and environments.
The SDK ships with plugins for LangChain, OpenAI Agents SDK, Anthropic Agents SDK, AutoGen, CrewAI, Semantic Kernel, Microsoft.Extensions.AI and MCP tools, positioning ACS as a common governance layer for agent deployments.

Sources

Left100%

Will Microsoft's open-source agent standard prevent vendor lock-in, or will it steer the industry toward its own AI ecosystem?

As firms adopt Microsoft's AI control standard, could its centralized policies become a single point of failure for attackers to exploit?

With the EU AI Act now in effect, can a technical standard truly solve the legal accountability gap when autonomous agents cause harm?