A mysterious character named Elias Thorne keeps appearing in AI-generated stories. Researchers say safety tuning and recycled training data may be to blame. The phenomenon is now spilling into self-published books and online content.
Content creators and publishers using AI tools are encountering a strange pattern: large language models like ChatGPT, Gemini, and Claude frequently generate stories featuring a character named Elias Thorne, often cast as a lighthouse keeper, clockmaker, or librarian. This recurring motif is now surfacing across self-published books, YouTube videos, and even questionable news sites, raising questions about the origins and risks of AI-generated content flooding digital platforms.
Software engineer Daniel May first identified the surge in Elias Thorne stories earlier this year. He noted that Google Trends showed little interest in the name until late 2025, with a sharp increase in early 2026. May tested several chatbots, including Grok, Deepseek, and Gemini, using simple prompts like “tell me a story.” The results were strikingly similar: stories about lighthouses, clockmakers, or explorers, often starring Elias Thorne or similar characters.
In May, researchers Sil Hamilton and David Mimno from Cornell University’s Department of Information Science published a study analyzing 20,000 stories generated by ChatGPT, Claude, Gemini, and the Allen Institute for AI's chatbot. Their findings revealed that the same 11 words—including names like Elias, Mara, and Elara, and roles such as lighthouse keeper, clockmaker, and librarian—appeared in over 88% of the stories, regardless of the model. Unite.ai reported on the study soon after its release.
The researchers suggest that these recurring themes are a byproduct of safety and alignment tuning in language models. Hamilton explained that most models are related through shared or synthesized training data, often originating from OpenAI’s GPT-3.5 and the WildChat dataset. WildChat, which contains one million real ChatGPT conversations, includes 166 instances of the name “Elias,” many written in the familiar lighthouse narrative style. As newer models were trained on data generated by previous models, this style and character spread widely—much like a digital virus.
The Elias Thorne phenomenon has now moved beyond chatbots. May discovered the name appearing as the author of various books on Amazon, ranging from alternative medicine guides to YouTube algorithm manuals and psychological thrillers. Some of these works, he noted, could cause real harm due to misinformation. Elias Thorne also appears as a protagonist in fantasy novels and as a musical artist producing ambient albums. One Amazon author profile, complete with an AI-generated photo, is linked to a series of AI-generated “grift” books. The influx of such content has made it harder for librarians and readers to distinguish credible works from AI-generated slop, echoing challenges seen when AI-generated mushroom foraging books and error-filled novels began to overrun the platform.
The character has also infiltrated YouTube and low-quality content sites. On the channel Moments That Moved the World, a video tells the story of “83-year-old Sergeant Major Elias Thorne.” Other sites spin tales of Elias as a snake museum owner or the “wealthiest man in Ohio,” often portraying him as a tragic or misunderstood figure. While Elias is a real name and has appeared in human-written stories—such as a BBC children’s writing competition finalist—there is no evidence that these mainstream outlets have been compromised by AI-generated content.
Researchers believe the prevalence of Elias Thorne and lighthouse stories is linked to the way model developers align outputs for safety. Hamilton said that many stories in the WildChat dataset were not safe for work, leading models to favor a narrow set of “safe” stories during alignment. As a result, the Elias Thorne narrative became a default fallback. The team plans to investigate this theory further.
Interestingly, the name Elias Thorne does have some historical echoes. In the 1980s trading card series Dinosaurs Attack!, Elias Thorne was a time-traveling mad scientist. Additionally, a real-life Elias Allen was a 16th-century clockmaker in London, which may explain the clockmaker motif in some AI-generated stories.
This pattern of AI-generated content crossing into mainstream platforms is not isolated. For example, recent reporting on AI-driven support vulnerabilities at Meta highlights how automation can introduce new risks for publishers and creators, especially when human oversight is limited.
