Generative AI Policy

Principles

• People remain responsible. The named authors are accountable for everything in their manuscript, including any text, citation or analysis produced with the help of a tool. Reviewers and editors are accountable for their reports and decisions.
• AI is not an author. A tool cannot be listed as an author or co-author, because it cannot give consent, hold copyright, take responsibility, or answer questions about the work.
• Confidentiality comes first. Unpublished manuscripts, reviews and decision letters are confidential and must not be given to tools that store or learn from what they receive.
• Be open about use. Use that goes beyond basic language help is disclosed, so that readers and editors can judge the work knowing how it was made.

Authors

What is allowed

• Using a tool to improve the language, grammar, clarity or readability of text the author wrote. This does not need disclosure, although authors may disclose it, and they remain responsible for the result.
• Using tools for purposes such as spell-checking, reference formatting or translation of the author's own text, again with responsibility remaining with the author.

What is allowed only with disclosure

• Using a generative AI tool to help with literature searching, outlining, summarising sources or drafting passages, or to analyse data. Authors must check that every statement, citation and quotation is real and correct against the original source, and must disclose the use as set out below. A tool's output must never be cited as a source of authority, and authors must not cite sources they have not read.

What is not allowed

• Listing a tool as an author, or giving a tool credit as a contributor.
• Creating or altering research data, results, figures or images with a generative tool without full disclosure in the manuscript. Where such use is disclosed, the editor decides whether it is acceptable, and may refuse work whose data or figures were generated by a tool.
• Presenting AI-generated text, analysis or citations that the author has not verified.
• Using a tool to disguise copied material or to generate a response to peer review that the author has not read and approved.

How to disclose

Add a short statement in the declarations (after the conclusion) that names the tool and version, says what it was used for, and confirms that the authors reviewed and take responsibility for the content. For example:

During the preparation of this work the authors used [tool name, version] to [purpose, for example "check the grammar of the introduction" or "suggest an outline for section 3"]. The authors reviewed and edited the output and take full responsibility for the content of the publication.

Reviewers

• A reviewer must not upload a manuscript, any part of it, or their review comments about it to a generative AI tool, because that would disclose confidential material.
• A reviewer must not ask a tool to assess the manuscript or write the review. The evaluation must be the reviewer's own.
• A reviewer may use a tool to improve the language of a review they have written themselves, as long as no part of the manuscript is entered into the tool.
• If a reviewer believes that the manuscript they are reviewing has been written or altered by a generative tool without disclosure, they tell the editor and give their reasons.

Editors

• Editors must not upload a submitted manuscript, a review, or a decision letter that discusses a manuscript to a generative AI tool.
• Editorial decisions and correspondence must reflect the editor's own judgement. A tool must not be used to make or draft a decision.
• Editors who use a tool for language only, on text of their own that has no confidential content, remain responsible for what they send.
• Editors may use tools for administrative checks that do not involve confidential content.

Detection and sanctions

The journals do not rely on AI-detection software as proof, because such tools are unreliable and can wrongly flag the work of authors who write in a second language. A concern about undisclosed use is raised on the basis of evidence, such as citations that do not exist, statements that the author cannot support, or the manuscript's own text, and the author is asked to explain.

Undisclosed or prohibited use is handled under the Publication Ethics procedure. In a published article, undisclosed use that affects the work can lead to a correction or a retraction as described in the Corrections and Retractions policy.

Keeping this policy up to date

Tools and practice are changing quickly. We review this policy regularly in light of guidance from COPE and the scholarly community, and each journal's Author Guidelines point to the current version.