Agent Governance Toolkit

Open-source runtime security covering all 10 OWASP agentic AI risks

Microsoft dropped the Agent Governance Toolkit on April 2nd — a seven-package, multi-language open-source system for bringing security and compliance to autonomous AI agents. It's the first toolkit to claim coverage of all 10 OWASP Agentic AI risks with deterministic, sub-millisecond policy enforcement. The toolkit includes zero-trust agent identity via Ed25519 credentials, execution sandboxing with 4-tier privilege rings, an MCP security scanner for detecting tool poisoning and typosquatting, and compliance automation mapped to the EU AI Act, HIPAA, and SOC2. It integrates with 12+ frameworks including LangChain, CrewAI, AutoGen, OpenAI Agents, and Google ADK — hooking into each framework's native extension points rather than wrapping them. Languages supported: Python, TypeScript, .NET, Rust, and Go. With 9,500+ tests and 662 stars at launch, this is unusually mature for a v1.0 open-source release. The timing is deliberate: EU AI Act compliance obligations for high-risk systems came into force in March 2026. This is Microsoft planting a flag in the agent security layer.

Panel Reviews

Ship

“9,500 tests and sub-millisecond policy enforcement out of the gate is impressive engineering. If you're shipping agents to production in a regulated industry, this is the governance layer you were going to have to build yourself anyway. Ship.”

Skip

“Microsoft open-source is great until the framework it hooks into changes its extension points. The 12-framework integration surface means this will rot fast without active maintenance. The OWASP coverage claim is marketing until independent auditors verify it.”

Skip

“Agent governance is the missing layer of the agentic AI stack. Every company deploying autonomous agents in 2026 will need something like this. Microsoft shipping it as open source is a power move — they're trying to own the standard.”

Skip

“This is pure infrastructure — no direct application to creative work. But if you're building tools that agents use, the MCP security scanner for tool poisoning is worth knowing about.”