Data poisoning manipulates training, fine-tuning, evaluation, or retrieval data so an AI system learns corrupted behavior or produces attacker-chosen outcomes. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.