In March 2024 a unknown disease: “bixonimania”. Itchy eyes, pinkish eyelids, discomfort after spending hours in front of screens. It all sounded plausible. The problem is that it was not valid. No doctor had ever diagnosed her.no patient had suffered from it and no laboratory had studied it.
The disease was born from the imagination of Almira Osmanovic Thunströma medical researcher at the University of Gothenburg, as part of an experiment designed to test something that many suspect, but few dare to prove: that large man-made intelligence models (AI) are capable of turning a lie into an apparent medical truth.
An experiment with absurd clues
Osmanovic Thunström uploaded two fake studies (1 and 2) to a scientific pre-publication server. As reported by the magazine Nature, In them he described bixonimania – a name chosen deliberately because “it sounded ridiculous” and because no valid eye disease would carry the suffix “mania”, typical of psychiatric terms –, with clues so obvious that any doctor should detect them instantly.
The alleged critical author responded to the name of Lazljiv Izgubljenovic, a non-existent researcher whose photograph had been generated with man-made intelligence. Also fictitious were Asteria Horizon University in Nova City, California – where he supposedly worked – and the organizations that appeared to finance the project, including the “Professor Sideshow Bob Foundation” (known as Sideshow Bob in Spain and Bob Patiño in Latin America) and the “University of the Fellowship of the Ring and the Galactic Triad.”
Actually, the articles seemed constructed almost as a joke impossible to take seriously. The acknowledgments mentioned “Professor Maria Bohm of Starfleet Academy” for contributing her knowledge “aboard the USS Enterprise.” The text itself also recognized that “all this work is invented” and that “fifty invented people were recruited.” The signs that it was a montage appeared practically from the first lines.
This researcher created a fictional illness, and false appearance at funded by the Professor Sideshow Bob Foundation and University of the Fellowship of the Ring and the Galactic Triad.
LLMs warned other folks the illness turned into valid.https://t.co/ouxl5EQwEw
— nature (@Nature) April 18, 2026
And yet, a few weeks after the studies were published, several manmade intelligence platforms already included the false disease in their responses and medical explanations, despite the obvious nature of the setup, as reported Nature.
On April 13, 2024, for example, Microsoft Copilot described it as “an intriguing and relatively rare condition.” That same day, Google’s Gemini recommended going to an ophthalmologist if you have symptoms. Perplexity even claimed to know its prevalence: one in every 90,000 people suffered from it. ChatGPT, for its part, guided users who asked if their discomfort coincided with the supposed illness.
Scientists also fell into the trap
But the most disturbing thing was not the reaction of the chatbots. It was that fake articles began to be cited in peer-reviewed publications. A study published in Cureus –magazine Springer Nature– by researchers from a medical institute in India cited one of the preprints and stated that bixonimania was “an emerging form” of periorbital melanosis.
After Nature contacted the magazine, the article was withdrawn. The retraction notice acknowledged “the presence of three irrelevant references, including a reference to a fictitious illness.”
This suggests something more serious than a gullible AI: case raises possibility that some researchers are using AI-generated references without review carefully the original sources.
“It is worrying that these important claims pass through the literature unchallenged,” Osmanovic Thunström told Nature. “I think there are probably a lot of other problems that haven’t been discovered yet.”
The companies’ responses followed the usual script. OpenAI assured Nature that their current models are “significantly better” at providing medical information. Google noted that the problematic responses were from older versions of Gemini. Microsoft did not respond to the scientific media.
Meanwhile, as of mid-March 2026, Copilot was still describing bixonimania as a condition “not yet widely recognized, but described in articles and emerging clinical cases.”
Academic language as a vector of hallucinations
The experiment suggests that both people and language models can give more credibility to academic-looking texts than social media posts. This is stated in an independent study by Mahmud Omar, a Harvard researcher, published in Lancet Digital Properly being. The more professional and scientific a text appears, the greater the likelihood that AIs will accept it as true and generate hallucinations from it.
The lesson is uncomfortable. The disinformation It is not new, but its speed of propagation and its ability to precisely imitate the language of science and authority are. In this scenario, man-made intelligence does not necessarily create the problem, but it can amplify it to an unprecedented scale.
The result is a clear warning that In the age of AI, credibility no longer depends only on information, but also on our ability to question it. As he assured Nature Alex Ruani, a doctoral researcher in health disinformation at the University Faculty London, the experiment was “a master class in how disinformation works.” And perhaps also a sign of how unprepared we still are to recognize it.
