AI tools generated 85+ flawed medical papers by students in 2023

Students queried tools including Elicit and Consensus to auto-generate literature reviews, statistical sections, and conclusions for PubMed-indexed submissions. Journal audits identified duplicated AI phrasing across unrelated specialties and citation chains that referenced nonexistent DOIs. Retraction notices rose 340 percent in targeted journals compared with 2022 baselines.

PubMed Central data through Q4 2023 show a 19-fold increase in papers listing medical-student first authors that contain LLM-typical artifacts such as hallucinated p-values and recycled methods text. No institutional review board records exist for 62 percent of these submissions. Mainstream reporting attributed the surge to individual misconduct and omitted the absence of watermark detection or authorship verification in preprint servers and editorial management systems.

The pattern replicates earlier waves of paper-mill output documented in 2019-2021 but now scales through zero-cost inference APIs. Journals lack mandatory AI-disclosure fields and similarity tools calibrated for synthetic text. Without cross-publisher detection mandates, the fraction of AI-generated medical literature is projected to exceed 8 percent of new PubMed entries by 2026.

Publishers have begun testing CRediT taxonomy extensions and LLM-output classifiers. Enforcement remains voluntary.