Scientists Came Up With a ‘Fake’ Disease. AI Convinced People It Actually Exists
In early 2024, a medical condition called bixonimania began appearing in AI-generated health advice, complete with symptoms, prevalence rates, and treatment suggestions, despite the fact that it had never existed in the real world.
The Experiment Behind the Illusion
The condition was deliberately invented by researcher Almira Osmanovic Thunström, who wanted to test whether AI systems would absorb and repeat false medical information. To do this, she created fake academic papers and uploaded them online, seeding the idea into the same ecosystem that large language models use to learn and generate responses.
Building a Convincing Fiction
The deception was layered with clues that should have raised alarms, including a fictional author, a non-existent university, and even references to Star Trek and The Lord of the Rings hidden in the acknowledgements. Some sections of the papers explicitly stated that the research was fabricated, yet the format still resembled legitimate scientific work closely enough to pass through digital filters.
AI Picks It Up and Runs With It
Soon after the fake studies appeared, major AI tools began treating bixonimania as a real condition, describing it as a rare illness linked to blue-light exposure and advising users to seek medical care. Some systems even estimated that one in 90,000 people might be affected, turning a fictional concept into something that sounded clinically grounded and statistically credible.
When Misinformation Becomes Self-Reinforcing
The problem escalated when the fake research began appearing in other academic work, including peer-reviewed studies that cited bixonimania as if it were legitimate. This revealed a deeper issue: not only were AI systems spreading the misinformation, but human researchers were also relying on AI-generated references without verifying their authenticity.
Why AI Was So Easily Misled
Experts say the format of the fake papers played a crucial role in deceiving AI systems, which are more likely to trust information that looks like formal medical literature. When text is structured like a clinical report or academic study, models are more prone to accept and elaborate on it, even if the underlying content is false.
A System Under Pressure
The experiment landed at a time when AI tools are increasingly used for medical advice, with surveys showing that about one in six adults rely on them at least monthly. As tech companies race to build AI-driven health platforms, the pressure to deliver fast, comprehensive answers can sometimes outpace safeguards designed to ensure accuracy.
Experts Sound the Alarm
Researchers studying misinformation warn that the bixonimania case is more than a curiosity, calling it a clear demonstration of how fragile information systems can be. “This is a masterclass on how mis- and disinformation operates,” one expert noted, emphasizing that failures in filtering and verification could have serious consequences in real-world health scenarios.
AI Begins to Correct Itself
Over time, many AI systems started to recognize the condition as fictitious, with newer responses flagging it as a made-up or fringe concept. Still, inconsistencies remained, with some models continuing to describe it as an “emerging” disease depending on how questions were phrased, highlighting the uneven progress in improving reliability.
A Warning About Trust in the AI Era
The rise and fall of bixonimania underscores a larger challenge: as AI becomes a central source of information, the line between verified knowledge and convincing fiction can blur quickly. The experiment shows that even obviously false data can ripple through systems and influence both machines and humans, raising urgent questions about how trust is built, maintained, and protected in an AI-driven world.