The Case of Elias Thorne, Imaginary Man AI Chatbots Are Obsessed With

A striking pattern emerged when researchers analyzed roughly 20,000 AI-generated stories from major large language models: the same handful of invented names and jobs dominates output. One fictional figure, Elias Thorne, repeatedly appears as a lighthouse keeper, clockmaker, librarian, explorer and protagonist across narratives, turning up in nearly two-thirds of the stories; together with a few other names and occupations, these tokens appear in about 88 percent of the sample. The character’s footprints extend beyond generated fiction into cheaply produced books, music listings, videos and even dubious health guides, yet investigators could not locate a clear original source for the name.

The explanation offered is structural rather than accidental: aggressive safety and copyright filtering narrows the legal training pool, and many current models are trained on material produced by earlier models, so invented content gets recycled across generations. That constrained, self-referential data ecology produces brittle, repetitive outputs that cannibalize their own inventions. Elias Thorne is therefore not just a quirky hallucination but a symptom of how alignment practices and dataset reuse can hollow out diversity and originality in chatbot storytelling.