Microsoft open-sources agent trust stack at Build 2026
Microsoft released an open, end-to-end trust stack for AI agents at Build 2026, introducing two open-source components designed to address safety and control challenges in agent deployment Microsoft Foundry.
Two Components: ASSERT and Agent Control Specification
The stack comprises ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) and the Agent Control Specification (ACS), both released as open-source tools Microsoft Foundry. ASSERT converts written policies into executable evaluation test suites, enabling organizations to validate agent safety and quality against defined standards Microsoft Germany. ACS standardizes deterministic safety and security controls at five critical checkpoints: input validation, model/LLM selection, state management, tool execution, and output filtering Microsoft Foundry.
Policy-Driven Evaluation at Scale
ASSERT addresses a core operational gap: translating high-level security and quality policies into repeatable test cases without manual engineering overhead. The tool enables teams to define compliance requirements, business rules, and safety constraints once, then execute them continuously as agents evolve. This approach shifts agent testing from ad-hoc validation to systematic, policy-aligned evaluation.
ACS takes a complementary approach by embedding control logic at deterministic points in the agent lifecycle. Rather than attempting to predict all possible agent behaviors, ACS uses guardrails at known decision boundaries—before the agent receives input, before it calls an LLM, before it executes tools, and before it returns output to users. This architecture allows operators to enforce rules without retraining models.
Industry Specification, Not Proprietary Lock-in
Both components are released as industry specifications rather than proprietary Microsoft products, signaling an effort to establish open standards for agent governance Microsoft Foundry. The open-source release aims to enable developers across platforms and cloud providers to implement compatible safety and control mechanisms.
Timing and Context
The announcement comes as enterprises increasingly deploy autonomous agents for customer service, finance, HR, and logistics workflows. As agent deployments scale, governance gaps have emerged: how to enforce consistent safety policies across heterogeneous agent architectures, how to audit agent decisions, and how to recover from failures. ASSERT and ACS are positioned as infrastructure-level solutions to these operational challenges.
Microsoft's trust stack release reflects a shift in the agent economy from prototype-to-production debate toward concrete governance tooling. Early adopters can now access the components via the open-source repositories and integrate them into existing agent frameworks and platforms.
