ChatGPT and Claude are inventing fake scientists to publish fake research

Researchers found that thousands of academic articles feature non-existent experts like Elena Vasquez or Marcus Chen. These individuals are mere hallucinations—digital phantoms birthed because ChatGPT, Claude, and Gemini have favorite names they default to when tasked with generating credible-sounding bibliographies.

The study highlights how these LLMs don't just pick random names, they build distinct “ghost ensembles.” For example, Claude has a weird soft spot for Elena Vasquez, while ChatGPT loves inserting Elara Voss into technical discussions. This behavior acts as a digital fingerprint, allowing researchers to guess which specific model hallucinated a paper by looking at the fake contributors list.

The scale of this phantom invasion is staggering, with over 1,500 articles identified on Zenodo that cite nonexistent journals and fake publication dates. While the current model-attribution trick works, it’s a race against time; as these fake papers get indexed by the internet, future models will train on them, eventually drowning the truth in a sea of AI-generated noise.

This marks the official start of the era where science is no longer about discovering facts, but about whose chatbot can hallucinate a more prestigious-sounding resume. It’s a perfect feedback loop of stupidity where the models are training themselves on their own creative lies, effectively turning the entire internet into one giant, automated game of telephone.

Source: 404 Media