A researcher-led investigation highlighted by Maxim Topaz’s team extends beyond paper-level mistakes, showing how fabricated references can become embedded in biomedical literature at scale. The study’s audit approach—reviewing millions of papers and tens of millions of citations—provides a quantitative baseline for educators, editors, and researchers trying to assess where failures are occurring in real workflows. The report describes hallucinations as a model behavior that can prioritize word patterns over accuracy, and it notes that the affected references may be introduced even when researchers use tools primarily for grammar and formatting support. Topaz’s account of a near-miss in his own submission underscores how quickly such errors can slip through expert review and editorial checks. By demonstrating a multi-fold rise in fake citations, the study effectively treats AI-assisted writing and verification as a compliance and scholarly-integrity issue, not merely a technical one. It also points to medicine’s reliance on citation chains that can propagate errors into trials, systematic reviews, and guidelines. For universities, the development points to potential needs for AI literacy training for faculty and journal submission systems that include citation verification guardrails when generative tools are used.