Microsoft's Newest Open-Source Project: Runtime Security For AI Agents

Microsoft has introduced the Agent Governance Toolkit, a new open-source project licensed under MIT, designed to provide runtime security governance for autonomous AI agents. The toolkit aims to address all ten agentic AI risks identified by OWASP, including goal hijacking, tool misuse, and identity abuse. It offers components like Agent OS (policy engine), Agent Mesh (secure communication), Agent Runtime (dynamic execution), Agent SRE (safeguards), Agent Compliance (governance verification), Agent Marketplace (plug-in management), and Agent Lightning (RL training governance). The toolkit supports Python, Rust, TypeScript, Go, and .NET, and is available on GitHub for those interested in leveraging Microsoft's approach to AI agent security.

(Source：Phoronix)