AI Firewall – Security Gateway and Reverse Proxy for LLM Traffic (github.com)

AI Firewall has been introduced as a robust security solution designed to act as a gateway between applications and Large Language Model (LLM) APIs including those from OpenAI and Anthropic. This innovative tool examines user prompts before they’re sent to LLMs and their outputs before reaching users, implementing essential features like data loss prevention (DLP) for sensitive information, prompt-injection filtering, and compliance tagging according to the EU AI Act. A streamlined dashboard built with React facilitates management and detailed monitoring, while the underlying architecture employs a low-latency Go proxy for efficient request processing. This development is significant for the AI/ML community as it addresses critical security challenges associated with utilizing LLM APIs, particularly the risks of sensitive data exposure and prompt injections, while ensuring regulatory compliance. The system relies on powerful detection mechanisms, including a deterministic keyword automaton and regex for known threats, though it currently outlines limitations with paraphrased or multilingual attacks. Plans to implement a more advanced ONNX transformer classifier aim to enhance detection capabilities further. The AI Firewall is designed to be source-available and free for limited use, underscoring a growing focus on safety and transparency in AI applications.