TL;DR
Microsoft's Agent Governance Toolkit is an open-source framework that provides policy enforcement, zero-trust identity, execution sandboxing, and SRE for AI agents, fully covering the OWASP Agentic Top 10.
Key features
Policy enforcement: Fine-grained control over agent actions (e.g., allow send_email, forbid drop_table)
Zero-trust identity: Track and identify each agent's behavior in multi-agent systems
Execution sandboxing: Run agent code in isolated environments to minimize security risks
Audit logs: Tamper-proof records for tracing policy violations and supporting regulatory compliance
Framework-agnostic: Integrate with any AI agent framework via pip install
When to use it
When deploying autonomous AI agents to production with security and governance requirements
When tracking and controlling each agent's behavior in multi-agent systems
When operating AI agents in enterprise environments requiring regulatory compliance and auditing