Patronus Protect is an on-device security layer for AI systems that detects prompt injections, prevents data exfiltration, and provides transparency into what AI agents actually do under the hood. We are building Patronus alongside our studies and jobs and want to contribute security tools and research back to the open source community.