Why LLM Chatbots Keep Inventing the Same Fictional Narrative

Multiple LLM-based chatbots including ChatGPT repeatedly generate nearly identical stories about a fictional lighthouse keeper named Elias Thorne, revealing that this fabricated narrative is embedded in their training data. This phenomenon exposes a critical vulnerability in AI systems: how false or fictional content can propagate through internet-scale training datasets and become embedded as learned patterns that models faithfully reproduce. The case exemplifies the intersection of data contamination, model hallucination, and the systemic challenge of validating information quality at scale.