BMJ Requires All Four ICMJE Criteria, and Names the One a Tool Cannot Meet

BMJ runs two separate policies that answer different questions: who counts as an author, and what has to be said about AI. Reading them together is what makes each one make sense.

Authorship. BMJ adopts the ICMJE’s four criteria in full: substantial contribution to the work’s conception, design, or data; drafting or critically revising it; final approval of the published version; and agreement to be accountable for all of it, including helping resolve questions about its accuracy or integrity. All four are required, not a majority of them. BMJ recognises “only natural persons” as authors, which rules out both organisations and AI tools in the same sentence.

Contributorship. Every submitted article needs a contributorship statement describing who did what, naming all author contributors and optionally non-author contributors such as those who “served as scientific advisors” or “collected data”. One named person is the guarantor, who “accepts full responsibility for the finished work”. Everyone named must agree to be included.

AI disclosure. “AI technologies will not be accepted as an author(s)”. Where AI was used, authors must disclose and describe it “to the best of their knowledge”, including in related work by the authors and in key cited sources. The disclosure goes in the contributor section — the same section the guarantor is named in — naming the technology, the reason for using it and the task it performed; use in the research itself is described more fully in the methods.

Verification. BMJ may use screening tools to detect AI-generated content, before or after publication, and authors remain responsible for accuracy, source attribution and avoiding plagiarism regardless of how the content was produced.

Placing the AI disclosure in the same section as the guarantor is a structural choice, not a filing convenience: it puts the statement about what a tool did next to the statement about who is answerable for the paper as a whole, so a reader checking accountability finds both facts in one place.

Requiring all four ICMJE criteria, rather than treating them as guidance, is what makes “only natural persons” a workable rule rather than a slogan: criterion four specifically requires agreeing to be accountable and to help resolve questions about accuracy, which is exactly the thing a model cannot do, whatever its output looks like.

The six-month review cycle on the AI policy, stated directly on the page, is a rare degree of explicitness among the publishers in this set about how current a policy is meant to be — worth noting because it tells a reader how stale the brief above can become before BMJ itself revisits the rule.

How it reaches a specific journal. These are BMJ group policies and apply across its journals, including specialty titles such as BMJ Case Reports; a specialty journal’s own instructions may add further detail without lowering the baseline set here.

Where a submission involves a consortium or study group, the group-authorship rule is worth reading closely: every individually named member still has to meet all four criteria themselves, which a large collaboration can overlook when assembling its author list.