JAMA draws a hard line on AI authorship to protect medicine from fabricated authority

Assistance is allowed when responsibility remains human

JAMA permits AI use in research when methods and reporting guidance are followed, and it permits assistance with manuscript preparation when the use is disclosed in the acknowledgment. Authors must check accuracy and accept responsibility for every part of the submission.

Basic grammar and spelling checks do not require disclosure. Translation can be used with disclosure and verification, although authors are warned about entering content into public tools.

The journal is closing routes for fabricated authority

The guidance says AI should not generate or format references because models can create citations with plausible authors and titles that do not exist. Standard reference managers remain acceptable, and authors are responsible for reference integrity.

AI drafting is not permitted for opinion manuscripts, letters, and online comments because the journal prioritizes human insight. AI-created or manipulated clinical images and multimedia are also barred unless they are part of disclosed formal research methods.

Confidential review stays outside external models

JAMA continues to prohibit AI use in peer review because entering an author's manuscript or a review summary into an external system can violate confidentiality. The same concern applies to unpublished patient, research, and commercial information.

The policy recognizes that disclosure alone cannot repair every fabricated reference, formulaic submission, biased image, or breach of trust. High-risk uses need categorical boundaries as well as transparency.

• Disclose the tool, version, manufacturer, dates, purpose, and human responsibility.
• Verify every reference with an authoritative bibliographic source.
• Keep confidential manuscripts and review material out of external models.
• Preserve original clinical media and document every permitted adjustment.