Microsoft Builds AI Governance Stack to Tame the Agentic Era

Microsoft used its annual Build developer conference to unveil a suite of tools aimed at making AI agents safer, more controllable, and easier to evaluate — a direct response to growing industry alarm over autonomous AI systems operating without adequate oversight.

At the centre of the effort is the Agent Control Specification (ACS), an open-source standard that lets development, compliance, and security teams define granular policies governing what an AI agent may do, what it must not do, when a human must approve an action, and what must be logged. The SDK ships with plug-ins for major frameworks including LangChain, Anthropic Agents SDK, OpenAI Agents SDK, and AutoGen.

Alongside ACS, Microsoft launched Scout, an always-on agentic assistant built on the OpenClaw framework and available through the company’s early-access Frontier programme. Scout VP Omar Shahine said the assistant is designed to learn individual working styles over time, accumulating memories and skills that make it progressively more capable for each user.

To help developers test whether AI behaves as intended, Microsoft also released ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing). Sarah Bird, Chief Product Officer of Responsible AI at Microsoft, said reliable evaluation is essential to building trustworthy AI systems, noting that application-specific behaviour requires far more dimensions of testing than general benchmarks currently provide.