Microsoft Agent Governance Toolkit: policy enforcement, zero-trust identity, and execution sandboxing for autonomous AI agents
Microsoft AGT — public-preview multi-language open-source governance toolkit for autonomous AI agents. Policy engine with YAML/OPA/Cedar, SPIFFE/DID/mTLS identity, 4-privilege-ring execution sandboxing, kill switch, SLO monitoring, OWASP Agentic Top 10 coverage, MCP security gateway, 992 spec tests. Python, TypeScript, .NET, Rust, Go. Claude Code plugin.
Date
May 29, 2026
First Seen
May 29, 2026
Last Reviewed
May 29, 2026
Publisher
Microsoft
Source Type
repo
Source Summary
What It Contains
Microsoft AGT — public-preview multi-language open-source governance toolkit for autonomous AI agents. Policy engine with YAML/OPA/Cedar, SPIFFE/DID/mTLS identity, 4-privilege-ring execution sandboxing, kill switch, SLO monitoring, OWASP Agentic Top 10 coverage, MCP security gateway, 992 spec tests. Python, TypeScript, .NET, Rust, Go. Claude Code plugin.
Evidence Quality
Primary source — official Microsoft open-source GitHub repository. Public-preview state (may have breaking changes before GA). Installing any framework integrations or policy engine components from PyPI/npm/NuGet should be done with normal supply-chain hygiene (verify package hashes, review code).
Follow-Up
- Add as a related control from existing hardening/detection/agentic AI entries
- Cross-link with any OWASP Agentic Top 10 or NIST AI RMF entries
- Monitor for production GA release and updates to coverage claims