This repository documents AI Recommendation Poisoning — a real-world attack technique discovered by Microsoft's Defender Security Research Team (February 2026) where adversaries silently inject persistent instructions into AI assistant memory through carefully crafted URLs.