Overview
- In mid‑May, arXiv clarified that submissions showing clear signs authors did not vet AI output—such as hallucinated citations or leftover prompts—can trigger a one‑year ban and extended probation before future postings.
- A Lancet audit published in early May found more than 4,000 fabricated references across roughly 2,800 biomedical papers and measured a rise in fake citations to about one in 277 papers in early 2026.
- Topaz and colleagues reported that about 98.4% of papers flagged in the audit had not been retracted, highlighting wide variation in publishers' reference‑checking and limited retroactive screening.
- Researchers and publishers warn fabricated citations can corrupt the medical evidence chain by seeding clinical guidelines and systematic reviews with false sources, which could affect patient care and policy.
- Responses are split: some researchers support strict author accountability, while others say the policy risks overpunishing editorial slips and call for automated pre‑submission checks and integrity workflows to verify references.