Bixonimania and the epistemic fragility of artificial intelligence in medicine: lessons from a fabricated disease

The rapid integration of large language models (LLMs) into clinical and biomedical ecosystems has created unprecedented opportunities for information synthesis, decision support, and patient engagement. Alongside these advances, however, an increasingly concerning phenomenon—the amplification of medical misinformation through generative artificial intelligence (AI) systems—has emerged.1 The case of bixonimania, a completely fabricated disease that nonetheless circulated in AI-generated outputs as if it were a real clinical entity, serves as a particularly illuminating example.