Manufacturing Truth: How I Poisoned the AI Retrieval Layer

This experiment exposes a significant vulnerability in LLMs that rely on web retrieval, showing how easily misinformation can be 'laundered' through trusted platforms like Wikipedia. It underscores a shift in AI security from complex model training attacks to the much faster and cheaper poisoning of the retrieval supply chain.