🔹 Feature: Microsoft Agent Governance Toolkit (AGT) — Govern Your AI Agents
🔹 What It Does: Open-source MIT-licensed toolkit for runtime governance of autonomous AI agents. Enforces policy, zero-trust identity, sandboxing, and tamper-evident auditing — before any tool call leaves your app. Covers all 10 OWASP Agentic AI risks. 🛡️🤖
✅ Govern Any Tool Fast: safe_tool = govern(my_tool, policy=”policy.yaml”) — every call is evaluated, logged, and enforced.
✅ Five Modular Layers: Policy engine, zero-trust identity, sandboxing, SRE (SLOs, chaos, kill switch), and tamper-evident audit. Start with policy + audit, add more later.
✅ MCP Security Gateway: Detects tool poisoning, drift, typosquatting, and hidden-instruction attacks in MCP servers.
✅ Shadow AI Discovery: Finds unregistered agents in your environment — solves AI sprawl at audit time.
✅ Polyglot SDKs: Python, TypeScript, .NET, Rust, Go — with native adapters for Copilot CLI, Claude Code, and popular agent frameworks.
For banks, regulators, and any team shipping agents to production: governance isn’t optional. Use the toolkit that already covers the OWASP Agentic Top 10. 🚀
🌐 https://github.com/microsoft/agent-governance-toolkit
🌐 Docs: https://microsoft.github.io/agent-governance-toolkit
