Microsoft's Agent Governance Toolkit Gets Its First Community Audit and the Security Stack Matures

Tools

Microsoft Agent Governance Toolkit: Runtime Security for Autonomous AI Agents

This is the most complete open-source answer yet to the question of what sits between your agent framework and production actions. Sub-millisecond policy enforcement, zero-trust identity, execution rings, and automated compliance mapping to EU AI Act and HIPAA -- all in one MIT-licensed monorepo that hooks into LangChain, CrewAI, OpenAI Agents SDK, and a dozen others without rewrites.

Agent Governance Toolkit v3.0.1 Ships Rust and Go SDKs, SBOM Signing, and 7 New Tutorials

v3.0.1 adds Rust and Go SDKs, SBOM generation in SPDX and CycloneDX formats with Ed25519 artifact signing, and expands fuzz targets to 7 -- covering prompt injection, MCP scanning, sandbox, and trust scoring. The dependency monitoring now covers 13 ecosystems, which matters if you care about supply chain risk in agentic pipelines.

Agent Governance Toolkit GitHub Repo: Full Architecture and OWASP Agentic Top 10 Mapping

The README lays out exactly why prompt-based safety fails in production -- 26.67% policy violation rate in red-team testing versus 0% for kernel-level enforcement. The capability breakdown and per-language SDK feature matrix makes it easy to scope what you actually need to adopt incrementally.

Analysis

Help Net Security: Why Agent Governance Infrastructure Has Lagged Agent Autonomy

Solid third-party breakdown of the toolkit's architecture for a security-focused audience. The framing -- OS kernels solved this for processes, service meshes solved it for microservices, now we need the same for agents -- is the clearest explanation of why this approach makes sense structurally.

The MCP Security Scanner Is the Toolkit Feature Nobody Is Talking About Yet

As MCP becomes the standard connection layer between agents and tools, the toolkit's MCP scanner for tool poisoning and typosquatting is going to matter more than most of the other features combined. Supply-chain attacks on agent tool definitions are the next obvious attack surface and this is the first production-grade scanner targeting it.