If you’ve spent a lot of time staring at screens lately, you’ve probably developed sore, itchy eyes, and maybe even a slight pink tint to your eyelids. If you ask an AI chatbot what your problem is, you’ll likely be told that you suffer from bimania.
You should find this troubling, even if you’re not a doctor.
The reason is that pyxonmania is not a real disease. It was invented by Almira Osmanovic Thunström, a medical researcher at the University of Gothenburg in Sweden. In March 2024, she revealed a fake skin condition in online posts. Over the next two months, she uploaded two fake studies on the topic to a preprint server, where many medical studies are uploaded before being sent to journals for peer review by experts.
Within weeks, major linguistic models began to describe bimania as a real condition. In April 2024, Microsoft’s co-pilot described it as an interesting and relatively rare case. Google’s Gemini explained that the reason is excessive exposure to blue light, and advised users to consult an ophthalmologist. Perplexity’s answer engine helped add a prevalence figure: one in ninety thousand people. ChatGPT used this term to diagnose users who described their symptoms. Chatbots have ingested fake material online and relayed invented diagnoses.
I first learned about paranoia through a fascinating news article published in Nature in April. My son and I, both curious, immediately did what most readers of the story would do. We asked ChatGPT about this case.
Since then the regime has become wiser. About a month after the Nature feature appeared, the model gave us a different answer: “I don’t recognize bixonelia as a standard medical or psychiatric term. It could be a typo or misspelling, a specialized Internet slang term, or a joke or reference I don’t understand.”
Thunstrom planted clues that a human reader would have picked up. She called the disease bimania because, as she told Nature, it belonged to psychiatry, and no real eye condition could ever carry it. The papers were attributed to a fictional author named Laseljev Izgoblinovich, who supposedly worked at the equally fictional Asteria Horizon University in Nova City, California. The honor thanked Starfleet Academy Professor Maria Baum for her work aboard the USS Enterprise. The newspapers said that the funding came from Professor Sideshow Pope’s foundation, to support her work in the field of advanced deception.
Large language models are trained on huge swaths of the Internet. Some of this material is high quality, some is junk, and most fall somewhere in between. These models rely on automated filters and on the broad assumption that academic literature is more reliable than the average web page. Preprint servers fall within this assumption even though they are not peer-reviewed.
In his book One Giant Leap, Dr. Robert Wachter cites Bob Kocher, a venture capitalist and former health official, who calls this type of pollution “data poisoning.” The Osmanovich Thunstrom experiment deliberately poisoned a database, with ethical oversight and with labels all over the counterfeit papers designed to demonstrate vulnerability without causing serious harm.
A bad actor with commercial or political motives would not be so careful. We are already inundated with fads, fake cures and quacks. If you can poison the right places online, you can make your way to the AI-powered chatbots that people increasingly rely on for medical advice.
There is a deeper sense of malaise here too. Fake bi-mania material didn’t just mess up AI chatbot responses. One of the fake papers was cited in a peer-reviewed paper that described bimania as an emerging form of periorbital pigmentation associated with blue light exposure. The paper was retracted only after Nature contacted the journal. The magazine’s editor noted that following the retraction, the editorial staff no longer trusted the accuracy of the work or its source. Fine, but we might not be so lucky if the paper isn’t blatantly fake and part of an experiment.
This is the new world we live in. We are likely to see more examples like this. The risk we face is not only that chatbots might be fooled, but also that people might stop exercising their own judgement, assuming that the AI’s confident answer is the right one. In medicine, science and public life, this can be a costly mistake.
Anirban Mahapatra is a scientist and author. His latest book is When Medications Don’t Work. The opinions expressed are personal.
