Over 90% of medical researchers already use generative networks when drafting manuscripts, yet academic compliance is drawing a hard line on algorithmic trust. Editors at the JAMA Network reviewed 105,538 submissions across its 13 journals and established strict operational boundaries. The reasoning is purely pragmatic: evidence-based medicine must be insulated from model hallucinations, clinical data distortions, fabricated citations, and diluted author accountability.
The publishing group enacted an outright ban on AI generation for Opinion columns, letters to the editor, online comments, and formal author rebuttals to reviewers. Reviewers themselves are strictly prohibited from feeding confidential manuscripts into external LLMs due to privacy concerns. Fabricating or altering diagnostic images—ranging from micrographs to CT and MRI scans—is completely off-limits unless the algorithm's performance is the core subject of the study. Authors may still use language models for copyediting and translation, but only with a complete audit trail: exact platform versions, prompts, query dates, and intended tasks must be disclosed. Legal and academic liability remains entirely with human authors.
While researchers have rushed into generative tools without guardrails, voluntary disclosure rates in submissions crept to just 5.97% over the past year. For executive leadership and the MedTech sector, JAMA's decision sets a clear compliance precedent: in highly regulated domains, the cost of an AI hallucination is prohibitive, and the illusion of autonomous algorithmic expertise inevitably collides with strict legal liability.