Active verification of AI input in articles

Clark discusses The BMJ’s AI policy.1 I think academic publishing is in a deep crisis. Virtually any neural network can quickly bypass standard quality filters. These models easily adjust text length, mimic flawless academic style, and format bibliographies without errors.The current strategy adopted by publishing houses—relying on superficial declarations—seems ineffective. They simply ask authors to disclose usage—something along the lines of “Yes, I used ChatGPT.”But this rests solely on the honour system. Recent data on publication ethics confirm that such declarations are often futile: they cannot stop covert plagiarism, detect hidden data hallucinations, or verify the actual authorship of an article.234We must move away from the outdated “take our word for it” approach and implement tools for active verification.Cardiology offers an excellent benchmark here. Flagship journals have long maintained a zero tolerance policy regarding the transparency of raw data. Publishing a clinical trial requires the submission of anonymised patient datasets.56…