Data poisoning is the deliberate contamination of training or retrieval data to corrupt AI behavior: planting examples that create backdoors, biasing models aga
The deliberate contamination of training or retrieval data to corrupt AI behavior: planting examples that create backdoors, biasing models against targets, or seeding documents that hijack RAG answers.
Because the RAG-era variant needs no training access, just the ability to get malicious content into a corpus the system retrieves from, like the public web, shared drives, or reviews. Defenses include source vetting, content sanitization, and anomaly monitoring.