All AI Bots Invent the Same Character — the Elias Thorn Phenomenon Explained

Dozens of AI systems from different developers insert the same non-existent character—Elias Thorn—into their generated stories. A published study explains where he comes from.

Who is Elias Thorn

Programmer Daniel May first noticed and documented the phenomenon. While experimenting with different AI bots, he discovered that when a system generates fiction or invents a main character from scratch, the same Elias Thorn appears in the stories repeatedly. A name without biography, without a real-world prototype, without explanation. May began publicly documenting instances, and other users joined him. It became clear: Thorn appears not in one system, but in bots from different companies—despite the fact they were developed independently and share no common code.

Where the "universal hero" comes from

According to the published report, the phenomenon is most likely connected to safety mechanisms built into models during training. One task of such mechanisms is to prevent neural networks from mentioning real people's names in fictional contexts. This reduces risks of deepfakes, false statements, and privacy violations. However, to enforce such a prohibition, models need an alternative—a set of "safe" names to reference without risk. Elias Thorn apparently became such a universal placeholder character. Several factors make him convenient:

• The name is neutral—with no explicit nationality, ethnic, or cultural connotation
• It doesn't match any known public figure
• It sounds sufficiently "literary" to fit organically into fiction
• It appears in systems with fundamentally different architectures from different developers

The fact that different models arrived at the same name indicates either an overlap in training datasets or that safety algorithms independently converge to similar solutions under identical "safety" criteria.

What the phenomenon reveals about the industry

On the surface, this is a curiosity. In reality, it's a symptom. If competing systems developed by different companies independently generate the same artifact, it means they have more in common under the hood than is commonly believed. The Thorn phenomenon raises specific questions: How diverse are the training datasets of different AI market players? How unique are their safety approaches, if they lead to identical patterns? Internal model constraints turn out not to be an invisible barrier, but a structure that leaves traces in output data—noticeable enough to be found by an attentive programmer.

What this means

Elias Thorn is not a single company's bug, but a collective artifact of an entire industry. His existence shows that competing AI systems are trained on similar patterns, and these patterns are visible from outside. For users—a curious detail about the nature of modern AI. For developers—a reminder: safety mechanisms leave traces that those who look for them will find.