On this question there is no live debate among the standards bodies that set authorship policy: a generative AI tool or system cannot be listed as an author on a scholarly work. The International Committee of Medical Journal Editors (ICMJE), the Committee on Publication Ethics (COPE), and every major scholarly publisher whose policy CASRAI reviewed for this guide (Elsevier, Springer Nature, PLOS, Wiley) converge on the same rule, and the same underlying reason: authorship carries accountability, and an AI system cannot be held accountable.
This guide covers what that consensus actually says, why the reasoning holds across every organization that has stated a position, and how it differs from the separate (and still-evolving) question of how to disclose AI assistance in a manuscript — for the practical mechanics of writing a disclosure statement, see CASRAI’s AI-assistance disclosure guidance rather than duplicating it here.
The short answer
No major journal, publisher, or research-integrity body currently permits an AI tool — ChatGPT, Claude, Gemini, or any other generative system — to appear in the author byline, be listed as a co-author, or be credited via a contributor taxonomy such as CRediT. This has been the case since the first wave of policies in 2023, and every subsequent policy update CASRAI has reviewed has reaffirmed rather than revisited it. AI assistance is a disclosure matter, not an authorship matter.
Why authorship requires accountability an AI cannot bear
The reasoning is not aesthetic or political — it follows directly from what authorship is defined to mean. ICMJE’s authorship criteria require, among other things, that an author:
- made a substantial contribution to the work,
- participated in drafting or critically revising it,
- gave final approval of the version to be published, and
- agrees to be accountable for all aspects of the work, ensuring questions about its accuracy or integrity are appropriately investigated and resolved.
An AI system cannot satisfy the fourth criterion, and this is the one every policy statement singles out. Accountability presupposes a legal and ethical agent capable of being questioned, of standing behind a claim, of holding or assigning copyright, and of facing consequences (correction, retraction, institutional sanction) if something is wrong. A generative AI tool has none of these properties — it cannot be deposed, cannot consent, cannot be disciplined, and cannot meaningfully “approve” a final manuscript in the sense the criteria intend. CASRAI’s Dictionary formalizes this as the AI as author term (the disallowed practice) and its rationale as author responsibility for AI use (the principle that stands in its place): human authors remain fully responsible for accuracy, integrity, and originality regardless of which portions of a manuscript an AI tool helped draft.
The ICMJE position
ICMJE first added explicit guidance rejecting AI authorship in its 2023 update, and the current ICMJE Recommendations (updated January 2026) carry a dedicated section, “V. Use of Artificial Intelligence in Publishing,” with a subsection on use of AI by authors. ICMJE’s own language is direct: chatbots such as ChatGPT “should not be listed as authors because they cannot be responsible for the accuracy, integrity, and originality of the work,” and “humans are responsible for any submitted material that included the use of AI-assisted technologies.” Authors must describe how AI tools were used, both in the cover letter and in the manuscript itself (methods or acknowledgements, depending on what the content affected). CASRAI’s Dictionary tracks this specific policy point as AI co-authorship rejection (ICMJE 2023).
The COPE position
COPE’s Position Statement on Authorship and AI Tools, first published in 2023, reaches the same conclusion from the same underlying logic, framed slightly more broadly than ICMJE’s clinical-journal context: authorship requires accountability, informed consent, and an intellectual contribution an AI tool cannot make. Because AI tools are not legal entities, COPE notes they also cannot assert (or fail to disclose) a conflict of interest, and cannot manage copyright or licensing agreements attached to the work — both of which are ordinary incidents of being a named author. COPE’s guidance is that all AI tool use (text generation, image creation, data analysis, code assistance, translation) must be disclosed, typically in the acknowledgements or methods, naming the specific tool and version and describing what it was used for. Routine tools — spelling and grammar checkers, standard statistical software — are generally treated as exempt from disclosure, the same distinction CASRAI’s own generative-AI disclosure statement term draws. COPE is a standards and advisory body rather than an enforcement one (see COPE (Committee on Publication Ethics) for how it relates to individual journal editors, who retain the actual decision-making authority).
Where publishers converge
Individual publishers’ editorial policies are not bound by ICMJE or COPE — journals can and do set their own rules — but every major publisher CASRAI checked directly for this guide has adopted a policy that lands in the same place, usually citing ICMJE and/or COPE explicitly:
| Organization | Position | Where stated |
|---|---|---|
| ICMJE | AI tools cannot be authors; use must be disclosed in the cover letter and manuscript | ICMJE Recommendations, Section V.A (updated January 2026) |
| COPE | Authorship requires accountability, consent, and intellectual contribution AI cannot provide; all AI use must be disclosed | COPE Position Statement on Authorship and AI Tools (2023) |
| Elsevier | AI tools “should not be listed as an author or co-author, nor cited as an author,” because authorship implies responsibilities only humans can hold | Elsevier, “The use of AI and AI-assisted technologies in writing for Elsevier” |
| Springer Nature | LLMs “do not currently satisfy our authorship criteria” because accountability “cannot be effectively applied” to them; use must be documented in Methods | Springer Nature AI editorial policy |
| PLOS | AI tools “cannot be credited as authors or contributors”; contributions must be reported in Methods or Acknowledgements | PLOS editorial policy on AI |
| Wiley | “Only humans can be listed as authors,” consistent with COPE’s position statement; AI use must be fully disclosed at submission | Wiley Generative AI Guidelines for Authors |
The consistency is the point: this is not one journal’s house style, it is the near-universal position of the organizations that set editorial policy across scholarly publishing. If a specific journal’s author guidelines are silent on the question, the safest working assumption is that the ICMJE/COPE position applies by default — but always check the individual journal’s current instructions to authors, since disclosure placement and exact wording requirements do vary publisher to publisher.
AI authorship vs. AI-assistance disclosure — a different question
“Can AI be an author?” and “how do I disclose that I used AI?” are related but distinct questions, and conflating them is a common source of confusion. The authorship question has a settled, near-unanimous answer (no). The disclosure question — what counts as AI use significant enough to require disclosure, where in the manuscript it goes, what a compliant disclosure statement looks like, and how placement differs by journal — is a practical, evolving area with real variation across publishers. CASRAI’s AI-assistance disclosure guidance covers that ground directly, including sample disclosure statements and what does and doesn’t require disclosure; this guide focuses on the authorship question specifically and defers to that page for disclosure mechanics rather than duplicating them.
What to do instead: disclosure, acknowledgement, and CRediT
Since an AI tool cannot be an author, substantial AI involvement in a manuscript is handled through three separate mechanisms, none of which is the author byline:
- Disclosure in the methods or acknowledgements — naming the tool, version, and what it was used for. See Acknowledgement (vs authorship for AI) and, for consolidated disclosures covering multiple uses across a manuscript, generative-AI disclosure statement.
- CRediT contributor roles for the humans who directed, verified, or edited AI output — not for the AI itself. A researcher who used an AI tool to draft code, then reviewed and tested every line, can legitimately claim the Software or Validation role for their own verification work. CASRAI’s CRediT for AI-generated content: where the line is covers this in more depth — how to represent human verification and oversight of AI-assisted contributions within the existing 14 CRediT roles, and CASRAI’s CRediT author statement guide has worked examples of contributor statements.
- Continuing, undiminished author responsibility — every claim, citation, and figure in a submission remains the named human authors’ responsibility whether or not AI helped produce it. This is the flip side of the accountability argument: it’s precisely because AI cannot be accountable that the human authors’ accountability cannot be reduced by using it.
A closely related but separate question — how AI tools may or may not be used by peer reviewers, as opposed to authors — is covered by CASRAI’s AI in peer review dictionary term; the confidentiality concerns there (uploading an unpublished manuscript to a third-party AI tool) are distinct from the authorship question this guide addresses.
Frequently asked questions
Can I list ChatGPT, Claude, or another AI tool as a co-author?
No. Every major publisher and standards body CASRAI reviewed for this guide explicitly prohibits it, on the grounds that authorship requires accountability an AI tool cannot hold. Use disclosure instead — see the table above for how each organization phrases the rule.
What if I only used AI for minor copyediting?
Most publisher policies (Springer Nature and PLOS explicitly; the pattern is consistent elsewhere) distinguish substantive AI use from routine tools like spell-checkers and grammar assistants, which generally don’t require disclosure. Where the line falls for your specific use case is exactly the question CASRAI’s AI-assistance disclosure guidance is built to answer.
Does this rule apply to preprints, not just peer-reviewed journals?
The rule itself (AI cannot be an author) is a near-universal editorial and ethical position, not a peer-review-specific one, so the same logic applies to preprint servers. CASRAI’s Dictionary notes a real example: a preprint that listed an AI system as co-author before the 2023 consensus emerged was subsequently flagged as non-compliant once the field’s position settled — see AI co-authorship rejection (ICMJE 2023).
Is CRediT itself a way to give an AI tool credit without calling it an author?
No — CRediT is a taxonomy for human contributors, and the same accountability logic that excludes AI from authorship excludes it from CRediT roles too. What CRediT can do is capture a human’s verification or oversight work when they use an AI tool as part of a contribution. See CRediT for AI-generated content: where the line is for how that distinction plays out role by role.
Who decides if my journal’s policy is unclear?
The editor is the actual decision-making authority for any individual journal — COPE and ICMJE set widely-adopted standards, but neither is an enforcement body for a specific submission. If a journal’s instructions to authors don’t address AI explicitly, the ICMJE/COPE position is the safest default assumption, but check with the editorial office directly for anything ambiguous or high-stakes.
US copyright law reached the same conclusion by a different route, and several years later. In March 2026 the Supreme Court declined to review the holding that human authorship is a bedrock requirement of copyright — a question ICMJE, COPE and publishers had already settled for themselves through authorship criteria and disclosure policy.
