Sentinel Protocol: Open-Source AI Firewall for LLM Security

Sentinel Protocol is presented as an open-source AI firewall designed to protect applications from vulnerabilities when interacting with Large Language Models (LLMs). The protocol operates as a local proxy, intercepting requests and responses between the application and LLM APIs such as OpenAI, Anthropic, and Google Gemini. Its primary function is to enforce security policies, preventing the leakage of Personally Identifiable Information (PII) and mitigating injection attacks.

The protocol's architecture includes several key components. On the input side, it scans for over 40 types of PII, employing a neural injection classifier, regex, and semantic similarity analysis for layered defense. It also features MCP poisoning detection, loop detection, and a deception engine. On the output side, it classifies toxicity, code execution, and hallucination signals, and performs real-time PII redaction and stego exfil detection. Governance features include OWASP LLM Top 10 coverage, MITRE ATLAS threat attribution, and a JSONL audit log.

The protocol boasts impressive performance metrics, including low proxy overhead and zero cloud dependency. It includes formal verification specs and an AI Bill of Materials (AIBOM) generator for compliance. The creator emphasizes the importance of running the protocol locally to maintain data privacy and control. The project is available on GitHub and npm, inviting feedback from developers.

Transparency Disclosure: This analysis was prepared by an AI language model to provide a comprehensive summary and strategic insights from the provided source text. The AI model has been trained to avoid hallucinations and ensure factual accuracy, but human oversight is recommended for critical applications.