Hey Product Hunt, we’re the team behind Patronus, building AI security in Regensburg, Germany. Over the past months, we’ve been working on specialized security models for prompt injection and other AI security tasks. We released much of that work openly, and our model family has now crossed 30,000 downloads on Hugging Face. Today we’re turning that work into something developers can plug directly into their AI products: Patronus Scanner. The idea is simple. Before your model or agent consumes untrusted content, scan it. Patronus Scanner works with text, public URLs, documents and MCP servers. It detects prompt injection and data leakage before the content reaches your AI system. You can use it through REST, Python, TypeScript and Rust SDKs, our CLI, Remote MCP, or native integrations for coding agents like Codex and Claude Code. We also built a browser Playground, so you can try the scanner immediately without setting anything up. Playground scans don’t use your quota, and you can make 100 API scans per day without an account. We started Patronus because agents increasingly consume content they did not create themselves: websites, documents, tool results and MCP descriptions. Every one of those inputs can become part of the model context, so we think the security boundary needs to move there as well. We’d especially love feedback from people building AI apps and agents: What untrusted input worries you most today, websites, files, tool outputs, MCP servers, or something else? Thanks for checking it out, Dominik and the Patronus team

Patronus Scanner

API to scan AI inputs for prompt injection and data leaks

API to scan AI inputs for prompt injection and data leaks