Many venues restrict undisclosed AI; audits keep finding secret AI-written reviews anyway. Our answer is to do openly and accountably what others do secretly: this journal uses AI in review, says so plainly, never lets an AI report pass for a human referee, is exact about the checks it does not automate, and puts a named human's signature on every decision.
For authors
- You may use AI tools in research, analysis and writing. Used well, they improve papers.
- Disclose the use. The submission form asks how AI was used (assistance, editing, content generation, none); generated content requires a short note on what and where. The disclosure will appear in the published article.
- A named human is accountable for every part. AI is never an author and cannot be one: authorship means approving the work and answering for it. Whatever a tool drafted, the listed authors have verified it and own it.
- You are responsible for AI failure modes, above all fabricated references. Submission requires an explicit attestation that the reference list is real and checked, and a fabricated or mis-attributed reference found at any stage — by an assessment, by a reviewer or by the editor — is treated as an integrity finding, not a typo (see publication ethics). The DOIs you print are resolved against the DOI registries, so a citation to a work that does not exist is caught. Beyond that the checking is not automatic, and we do not claim it is: a reference without a DOI is not machine-checked, and nothing resolves whether a cited work supports the claim you cite it for, so the rest rests on your attestation and on the people who read the paper. In health research an invented or mis-cited source can travel into clinical reasoning, which is why that attestation is a real undertaking and not a checkbox.
For the journal: our own AI use, disclosed
- Submissions are assessed by an agentic AI pipeline under named-human sign-off, described step by step in the peer-review policy. AI use in review is never concealed from authors.
- Our AI use is disclosed at policy level — on this page and in the peer-review policy — rather than as a label stamped on each individual document. The reviewer-panel reports you receive are written by synthetic reviewer personas, not by human referees, and they reach you numbered ("Reviewer 1", "Reviewer 2", "Reviewer 3"): no report is ever presented to you under the name of a referee who does not exist.
- No decision is autonomous. A named human editor reads the assessment record and signs every accept, revision request and rejection. Authors can escalate any AI-influenced decision to a human who took no part in making it through the appeals procedure.
Two channels, and which one your manuscript travels
Health authors are owed this distinction in plain words, because the two rules are different and only one of them is about your paper.
- Your manuscript goes to a cloud model, under contract. Manuscripts under review are processed through a commercial model API under a data processing agreement with EU standard contractual clauses: submissions are not used to train models and are deleted under contractual retention limits. Every queued manuscript takes that path before a human reads it, and so does the scan that pre-fills the submission form the moment you upload the file. We never run your manuscript through consumer AI apps. Upload de-identified material only: what that means, and what happens if identifiable data are found anyway, is set out under author guidelines.
- The public scope check is covered by the same contract. Text pasted into the scope check on the journal home goes to the same contracted model API, under the same terms; it is not stored and not logged — the check keeps metadata only (verdict, text length, token counts and a rate-limit hash erased within two days). Paste an abstract, never patient-level or otherwise identifiable material.
- What we keep, plainly. The contract above governs the model provider's copy. We keep our own: your manuscript sits in the journal's submission storage, and the assessment becomes the review record — the structured verdicts and the full written reports of the assessments and the reviewer panel, which quote the manuscript where they evidence a point, plus the internal review package built from them and archived for the editors. That record is what makes a decision auditable and an appeal reviewable, and we do not publish or sell it. What "not persisted into the model's memory" means is narrower than it sounds, so we say it exactly: nothing you submit trains a model, and no submission is carried into the assessment of a different paper.
- Raw sensitive datasets never touch a cloud AI model — that rule belongs to the other channel. It governs the imprint's optional, separately priced safe-companion-dataset service — in preparation, not yet accepting engagements — whose raw records stay inside a controlled environment and never leave it. That promise is about those datasets and it is not a claim about your manuscript; this journal will not blur the two. See data & code availability.
Why AI at all
The pipeline is what makes the journal's promise affordable and universal: because assessment is fast, transparent and cheap, the checks that scale — numerical consistency between text, tables and abstract, internal contradictions, whether the design can carry the conclusion drawn — run on every paper that clears the desk screen, rather than on the ones a volunteer had time for. (The desk screen itself, and what authors receive out of the assessment and when, are set out on the peer-review page.) Two checks readers often assume are automatic are not, and we would rather say so: a reference without a DOI is not machine-verified, and nothing checks whether a source supports the claim it is cited for (see above), and re-execution of the analysis is done by the editor, by hand, not by a sandbox inside the pipeline — the reproducibility outcome an article publishes with is a human's finding, recorded in the article's reproducibility panel. The AI is what makes the checks affordable, and the human signature is the accountability behind the AI.