Yet another experiment proves it's too damn simple to poison large language models

A security engineer demonstrated 'retrieval poisoning' by registering a $12 domain and adding a fabricated Wikipedia entry that caused multiple web-connected AI chatbots to assert a false championship. The writeup highlights three risk areas—search/retrieval ranking, possible inclusion in model training data, and compromised AI agents with tool access—and warns that simple SEO-style tactics can be scaled to produce significant misinformation and security risks for AI-driven systems.