LLM Guard & LeakGuard

A pair of AI-safety systems. LLM Guard is multi-stage middleware combining rule-based heuristics, classification models, and semantic techniques to detect prompt injection, jailbreaks, and sensitive-data exposure — producing explainable allow/flag/block decisions fit for enterprise audit. An output-validation pipeline flags hallucinated content, PII leakage, and policy violations before responses reach end users.

LeakGuard is a browser extension doing data-loss prevention at the edge: it intercepts input locally and blocks API keys, tokens, and PII from ever reaching external AI interfaces — processing fully on-device, zero data transmission.