JAMA Network has updated its guidance for authors’ use of artificial intelligence in journal submissions, advising against its use for bibliographic references, prohibiting its use to draft certain opinion and correspondence content, and restricting artificial intelligence–created or manipulated clinical media. 

The editorial, written by JAMA Network editors and published online August 10, 2026, in JAMA, addresses the expanding use of artificial intelligence (AI) in research and scientific publishing while emphasizing concerns about inaccurate or fabricated content and low rates of author disclosure.  

The updated guidance is intended "to promote and preserve accuracy and integrity of publication at a time when upholding the trustworthiness of the scientific record is more critical than ever," the editors wrote. 

One major change concerns bibliographic references. Authors are advised not to use AI, including large language models (LLMs), to generate, format, or otherwise manage references because the tools can produce inaccurate or nonexistent citations. The editors reported that JAMA Network has received submissions containing multiple inaccurate or nonexistent references and noted that frontier AI models can produce realistic-looking citations involving real authors and titles that appear to support a particular point but do not exist. Instead, the editors encouraged the use of standard reference managers and emphasized that authors remain responsible for checking all references for completeness, accuracy, and integrity. 

The updated guidance also prohibits using AI to draft Opinion manuscripts, Letters to the Editor, and online Comments. “We have seen marked increases in submissions of these types of articles, occasionally from the same authors, who may not have expertise in the topics they are addressing. These submissions are often formulaic and lack useful insights,” the editors wrote. AI also cannot be used to draft author responses to peer reviewer and editor comments.  

AI tools remain permissible for checking grammar and spelling and for translation after the initial text has been written by humans. Basic grammar and spelling checks do not require disclosure. Translation must be disclosed, and authors must understand the risks of placing manuscript content into an AI model and take full responsibility for any content changed by the tool. 

More broadly, use of AI, LLMs, chatbots, or similar technologies to assist with content creation, revision, or formatting must be disclosed in the Acknowledgment section. Required information includes the AI platform, program, or tool; version and extension numbers; manufacturer; dates of use; a brief description of how the technology was used and which portions of the manuscript were involved; and confirmation that the authors take responsibility for the accuracy and integrity of generated content. Incomplete disclosure or failure to disclose AI use detected during editorial review or after publication may result in rejection or post-publication corrective action. 

The guidance permits AI use in research when it is fully described in the Methods section, applicable reporting guidance is followed, and authors take responsibility for the content. AI-assisted literature searches also are permitted if their use is disclosed in the Methods section for research reports, or the Acknowledgment section for other manuscripts and authors verify the content for accuracy. However, presenting AI-generated data as original research derived from non-AI sources is not permitted. 

JAMA Network also strengthened its restrictions involving clinical images, illustrations, video, and audio, citing concerns about accuracy, clinical fidelity, and demographic representational bias. Clinical images and illustrations created or manipulated by AI cannot be submitted unless they are part of a formal research design or methods and the use is fully disclosed. The same restriction applies to AI-created or manipulated clinical video and audio. 

Uniform digital adjustments to brightness, contrast, or color remain permissible when applied to an entire scientific image, provided they do not selectively highlight, misrepresent, obscure, or eliminate clinical or scientific elements. Replacing a non-clinically relevant photographic background with a solid color also is permitted. Such modifications must be described in the figure legend, including the name and version of the AI or software used, and original unedited photographs must be provided for editorial review. AI-generated graphic visualization of research data or results also is permitted if the software name and version are identified in the figure legend. 

The editorial, written by JAMA Network editors, is a publication-policy statement rather than an original research study and therefore does not report a study population, statistical analysis, effect estimates, subgroup analyses, or quantitative study limitations.  

Roy H. Perlis, MD, MSc, reported advisor fees from Circular Genomics, Genomind, Alkermes, and Atella outside the submitted work. No other disclosures were reported.