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

A security engineer demonstrated how easily AI chatbots can be poisoned by fabricating false information—creating a fake Wikipedia entry and website claiming he won a non-existent 2025 card game championship. Multiple search-backed LLMs confidently reported this false "fact," revealing how they fail to verify source credibility. The exploit targets the retrieval-augmented generation (RAG) layer where AI systems search the web for supporting evidence.