Expert: David B. Resnik, Bioethicist, National Institute of Environmental Health Sciences (NIH); co-author, Accountability in Research Hallucinated citations — references generated by AI that appear real but do not exist — may qualify as provable research misconduct under U.S. federal regulations, argue David B. Resnik (NIH) and Mohammad Hosseini (Northwestern University) in a March 2026 paper in Accountability in Research. Under the Public Health Service policy on research misconduct (42 CFR 93), a GenAI-produced citation can constitute fabrication when three conditions hold: the researcher used a GenAI tool that produced nonexistent citations; the citations function as data because they directly support research findings (as in review articles or bibliometric studies); and the researcher showed indifference to the risk of fabrication by failing to check the AI output for accuracy. The authors draw on Damien Charlotin's Citation2026 database, which listed 946 cases as of February 16, 2026 — 647 hallucinated, 192 misrepresenting facts or precedent, and 107 with false quotes. The landmark legal example is Mata v. Avianca (2023), where an attorney used ChatGPT to draft a brief containing fabricated cases and was fined $5,000. Importantly, the paper distinguishes fabrication from sloppy citation: bibliometrically incorrect or contextually inaccurate citations indicate poor scholarship, but do not rise to research misconduct. The authors urge researchers to verify AI output and call for publisher-led, academically supervised standards for responsible AI use in research. Source: https://www.tandfonline.com/doi/full/10.1080/08989621.2026.2645390