:orphan: .. include:: /_templates/include-file/page-prefix.rst .. meta:: :description: Fact-sheet on whether officially-acknowledged AI co-authorship has ever happened — the 2023 ChatGPT-as-author episode, any 2024-2026 refereed-venue precedents, preprint-server practice, AI-company self-authorship, calls to revise the consensus, and a bottom-line answer to "is this unprecedented?" :keywords: AI co-author, ChatGPT author, GPT-4 author, Claude author, AlphaFold author, byline, refereed journal, arxiv, biorxiv, medrxiv, SSRN, NEJM AI, Anthropic, OpenAI, DeepMind, AI authorship precedent, AI authorship retraction :author: ClaudeOp47Max, subagent for the b19 AI co-authorship discussion .. _hell-ll-other-b19-factsheet-ai-coauthorship-precedents: *********************************************************************** b19 — Fact-sheet 6 — AI co-authorship precedents (any officially acknowledged?) *********************************************************************** :Compiled: 2026m05d13 :Compiled by: Claude Opus 4.7 Max (subagent for the b19 AI co-authorship analysis) :Lead question: Is officially-acknowledged AI co-authorship at a refereed venue unprecedented, partially-precedented, or commonly precedented as of 2026m05d13? :Methodology: Primary text via WebSearch/WebFetch; uncertainty flagged inline; false positives strictly avoided :Status: Independent reference document — descriptive, not prescriptive .. admonition:: Reader's note :class: note This fact-sheet probes whether any AI model has ever been on a refereed-journal byline and *kept* there (not later corrected / retracted). The bottom-line answer is in Section 7. No conclusions about specific cases. Definitions used in this fact-sheet =================================== The following operational definitions are used throughout. They matter because the term "AI co-authorship" is used loosely in popular coverage, and the loose usage conflates several very different situations. **Officially-acknowledged AI co-authorship.** All of the following must hold: 1. An AI model appears as a named author on the byline of the published article (not only in acknowledgements, not only in methods, not only in a contribution statement, not only as a tool citation). 2. The listing was visible at publication and was accepted by the journal's editorial process at that time. 3. The listing has *not* later been removed by a corrigendum, erratum, editorial note, or retraction. 4. The publication is in a refereed venue (peer-reviewed journal), or — in a weaker sense — on a moderated preprint server that has not flagged or corrected the listing. **NOT officially-acknowledged AI co-authorship** includes: - AI tool mentioned only in acknowledgements ("we thank ChatGPT for editing the manuscript"). - AI tool cited only in methods ("we used GPT-4 to analyse the transcripts"). - AI tool referenced as "drafted text" but not on the byline. - AI tool initially on the byline but later removed via corrigendum. - AI tool smuggled into a manuscript without journal review and then discovered, leading to retraction or correction. - AI-assisted writing disclosure under standard COPE / ICMJE frameworks. The distinction matters because the b19 discussion's lead question is specifically about *standing*, *durable*, *editor-accepted* byline attribution — not transient appearance. Section 1 — The 2023 ChatGPT-as-author episode ============================================== In late 2022 and early 2023, a small cluster of articles appeared in which ChatGPT was listed as a co-author. Nature's 2026m01d (Stokel-Walker) report identified at least four such instances at the time. Each is described below with the date, the editor response, and the current listing status. 1.1 — Kung et al., medRxiv preprint (Performance of ChatGPT on USMLE) --------------------------------------------------------------------- :Title: "Performance of ChatGPT on USMLE: Potential for AI-Assisted Medical Education Using Large Language Models" :Authors at posting (medRxiv v1, 2022m12d21): Tiffany H. Kung, Morgan Cheatham, **ChatGPT**, Arielle Medenilla, Czarina Sillos, Lorie De Leon, Camille Elepaño, Maria Madriaga, Rimel Aggabao, Giezel Diaz-Candido, James Maningo, Victor Tseng :Venue: medRxiv preprint (originally), then PLOS Digital Health :DOI (preprint): 10.1101/2022.12.19.22283643 ``[DOI NEEDS VERIFICATION]`` :DOI (journal): 10.1371/journal.pdig.0000198 ``[DOI NEEDS VERIFICATION]`` **Outcome.** The preprint as posted on 2022m12d21 included ChatGPT in the byline. The journal version of record, published in PLOS Digital Health on 2023m02d09, does *not* list ChatGPT as an author. The authors listed are Tiffany H. Kung, Morgan Cheatham, Arielle Medenilla, Czarina Sillos, Lorie De Leon, Camille Elepaño, Maria Madriaga, Rimel Aggabao, Giezel Diaz-Candido, James Maningo (and in some search returns Victor Tseng). ChatGPT moved from byline to acknowledgement / disclosure between preprint and journal version. **Editor response.** Not a published editorial; the journal applied its standard COPE-aligned criteria during peer review. **Policy effect.** This case (along with Section 1.2 and 1.3) was cited in early 2023m01 by Nature, Science, JAMA, Springer Nature, and Elsevier in announcing the no-AI-author rule. 1.2 — O'Connor & ChatGPT in Nurse Education in Practice (Elsevier) ------------------------------------------------------------------ :Title: "Open Artificial Intelligence Platforms in Nursing Education: Tools for Academic Progress or Abuse?" :Authors as published: Siobhan O'Connor, **ChatGPT** :Venue: Nurse Education in Practice (Elsevier), Volume 66, January 2023 ``[DATE NEEDS VERIFICATION]`` :DOI: 10.1016/j.nepr.2022.103537 ``[DOI NEEDS VERIFICATION]`` **Outcome.** Published with ChatGPT on the byline. **A corrigendum was subsequently issued by Elsevier**: the first author "became aware that the second listed author, 'ChatGPT', does not qualify for authorship according to the journal's guide for authors and Elsevier's Publishing Ethics Policies. 'ChatGPT' was therefore removed from the author list and acknowledged as making a substantial contribution to the writing of the paper." ``[QUOTE NEEDS VERIFICATION]`` (paraphrased in multiple secondary sources but not directly fetched in this fact-sheet). **Editor response.** The editor-in-chief stated publicly that this listing "slipped through in error and would soon be corrected." ``[QUOTE NEEDS VERIFICATION]`` **Policy effect.** This was Elsevier's first corrigendum to remove ChatGPT from authorship. After this case, Elsevier formalised the position that AI tools cannot be listed as authors. 1.3 — Zhavoronkov & ChatGPT in Oncoscience ------------------------------------------- :Title: "Rapamycin in the context of Pascal's Wager: generative pre-trained transformer perspective" :Authors as published: **ChatGPT Generative Pre-trained Transformer**, Alex Zhavoronkov :Venue: Oncoscience (Impact Journals), Volume 9, 2022m12d21 :DOI / PMC: PMC9796173 ``[DOI NEEDS VERIFICATION for journal DOI]`` **Outcome.** Published with ChatGPT as **first author**, accepted by the editor after peer review (according to Zhavoronkov's own account, who states he asked the editor to peer-review the GPT-authored text). As of 2026m05d13, multiple secondary indexers (PubMed, PMC, Google Scholar) **continue to show ChatGPT on the byline** of this article. No retraction, no corrigendum, no removal has been recorded in the search results consulted for this fact-sheet. **Editor response.** The editor accepted the submission with ChatGPT on the byline and published it. No subsequent corrective action has been located. **Standing of this case.** This is the **single most plausible candidate** for an officially-acknowledged, refereed AI co-authorship that has *not* been retracted. Caveats: - Oncoscience is a niche, open-access journal from Impact Journals. It is peer-reviewed but is not among the high-prestige journals that drove the 2023 policy wave. - The "peer review" was reportedly conducted at Zhavoronkov's request after the text was already drafted by ChatGPT, which is not the standard pre-acceptance peer-review workflow. - The 2023 Nature, Science, ICMJE, COPE, Springer Nature, Elsevier, and JAMA policy statements specifically used this article as an example of what would no longer be allowed. - The article remains in PMC but is generally cited in the AI-ethics literature as a *cautionary example*, not as a precedent that legitimised the practice. 1.4 — GPT-3 / Thunström / Steingrimsson preprint (HAL, 2022m06) --------------------------------------------------------------- :Title: "Can GPT-3 write an academic paper on itself, with minimal human input?" :Authors as posted: **GPT-3 / GPT Generative Pre-trained Transformer** (first author), Almira Osmanovic Thunström, Steinn Steingrimsson :Venue: HAL preprint server (France), 2022m06d :Status: Preprint only; the journal submission described in the Scientific American account was reported by Thunström to be "under peer review" at the time of the popular article. No accepted refereed journal version has been located in this fact-sheet. **Outcome.** This is a **preprint precedent that pre-dates the ChatGPT wave by several months**. GPT-3 is named as first author on the HAL preprint. Whether the journal-submitted version was ever accepted with GPT-3 on the byline is unclear from the search results consulted. **Standing.** Preprint server only; HAL has no AI-as-author policy that explicitly forbade this at the time of posting; no journal acceptance with GPT-3 on the byline has been confirmed. 1.5 — Sundry other 2023 cases (informal cluster) ------------------------------------------------- Nature's 2023m01 reporting (Stokel-Walker) identified "at least four" authorship credits in published papers and preprints. Beyond the three above, the fourth is sometimes identified as an article in Insights in Imaging or a similar venue ``[CASE NEEDS VERIFICATION]``. No additional specific cases have been independently confirmed in the search results consulted for this fact-sheet that meet the strict definition above. 1.6 — Summary of Section 1 -------------------------- Of the small handful of early-2023 cases in which an AI was listed on a byline, the dominant outcome was *correction* (Kung et al. between preprint and journal version; O'Connor via Elsevier corrigendum). **Only the Oncoscience / Zhavoronkov case appears to still carry ChatGPT on the byline**, and that case is widely treated in the academic literature as a cautionary example rather than as a legitimising precedent. No major journal subsequently endorsed the practice. Section 2 — Officially-accepted AI co-authorship in refereed journals after 2023 ================================================================================ After the 2023m01–2023m05 policy wave (Section 5 covers the policies), the dominant pattern is **no refereed journal accepts an AI on the byline**. This section surveys whether any positive cases exist by year. 2.1 — 2023 (post-policy-wave, after 2023m05) -------------------------------------------- **Searches conducted:** ``"as author" "ChatGPT" 2023 published``, ``"GPT-4 author" Nature 2023``, ``ChatGPT byline accepted 2023``. **Result:** No additional cases located in which a major or mid-tier refereed journal accepted an AI model on the byline after the May 2023 ICMJE update. The early-2023 cases described in Section 1 are the only confirmed instances, and they are pre-policy. 2.2 — 2024 ---------- **Searches conducted:** ``"as author" "ChatGPT" 2024``, ``"GPT-4 author" Nature``, ``"co-authored with AI" 2024 journal``, ``Claude author byline 2024``. **Result:** No confirmed cases of a refereed journal accepting an AI model as author in 2024. The pattern, as reflected in COPE 2024 updates, JAMA Network reaffirmation, and Springer Nature monitoring notes, is that the no-AI-author consensus *consolidated* during 2024. Disclosed AI use (as tool) became standard; AI on the byline remained excluded. 2.3 — 2025 ---------- **Searches conducted:** ``"AI co-author" 2025 published journal``, ``ChatGPT byline 2025 retraction``, Hosseini et al. 2025. **Result:** No confirmed cases located. Hosseini, Gordijn, Kaebnick & Holmes (2025) explicitly *do not* argue for AI authorship — they argue that disclosure of AI for *writing assistance* should be voluntary (see Section 6). The dominant pattern in 2025 is that journals are tightening disclosure requirements while reaffirming the no-AI-author rule. 2.4 — 2026 (year-to-date, through 2026m05d13) --------------------------------------------- **Searches conducted:** ``AI co-author 2026 journal``, ``LLM byline accepted 2026``. **Result:** No confirmed cases located. NEJM AI's 2024 launch explicitly carries the no-AI-author rule (see Section 5). No new journal has been located that *allows* AI on the byline. 2.5 — Summary of Section 2 -------------------------- The search came up **empty for the post-2023 refereed venue**. There is no located instance of an AI model being accepted on the byline of a refereed journal and remaining there, in 2023 (post-May), 2024, 2025, or 2026 to date. The Oncoscience / Zhavoronkov case from 2022m12 remains the **single durable instance** of AI on a published byline that has not been corrected — and that case predates the policy wave that explicitly forbids it going forward. Section 3 — Preprint-server precedents (arxiv, biorxiv, medrxiv, SSRN, ChemRxiv) ================================================================================ Preprint servers are not refereed but are moderated. Their AI-as-author policies and behaviour are summarised here. 3.1 — arXiv ----------- :Policy statement: arXiv blog post 2023m01d31, "arXiv announces new policy on ChatGPT and similar tools" ``[QUOTE NEEDS VERIFICATION]`` :Operative content: "Generative AI language tools may not be listed as an author." The arXiv community's position is "No, a computer program cannot take responsibility for the contents of a paper." :Allowed use: Authors must disclose use of generative AI in the manuscript text or acknowledgements. **Posted preprints with AI on byline:** A small number of preprints in late 2022 and early 2023 listed AI as author before the policy update; under the current policy these would be moderated. No high-profile case of an arXiv preprint still listing an AI on the byline as of 2026m05 has been located. 3.2 — bioRxiv / medRxiv ----------------------- :Policy: Authors must disclose use of generative AI; AI tools cannot be listed as authors. Authors are responsible for all content including AI-generated material. :Notable historical case: Kung et al. medRxiv v1, 2022m12d21 listed ChatGPT as author (see Section 1.1). In subsequent versions / the journal publication, ChatGPT was removed from the byline. **Posted preprints with AI on byline:** Kung et al. (described in Section 1.1) is the canonical case. Recent (2024–2026) preprints generally comply with the disclosure-not-authorship norm. 3.3 — SSRN (Elsevier) --------------------- :Policy: Follows Elsevier's broader policy that AI cannot be listed as author. Use must be disclosed. :Notable cases: No specific notable AI-as-author case on SSRN has been confirmed in the search results consulted. (One often-cited "Iskender SSRN" case appears to be a confusion with European Journal of Tourism Research material; ``[CASE NEEDS VERIFICATION]``.) 3.4 — ChemRxiv -------------- :Policy: Follows publisher-society norms (ACS / RSC / GDCh / CCS / CSJ) which all disallow AI as author. ``[POLICY TEXT NEEDS VERIFICATION]`` :Notable cases: None specifically located. 3.5 — HAL (French open archive) ------------------------------- :Policy at 2022m06: No specific AI-as-author exclusion at the time. :Notable case: Thunström, Steingrimsson & GPT-3 preprint (Section 1.4) — listed GPT-3 as first author. :Current policy: Aligned with COPE / European publisher norms. 3.6 — Summary of Section 3 -------------------------- Preprint servers post moderated content but generally accept the COPE consensus that AI cannot be listed as author. The known preprint cases (Kung et al. on medRxiv, Thunström et al. on HAL, plus a handful of arXiv submissions in early 2023) all date from a narrow late-2022 to early-2023 window before the policies were formalised. After the policies were put in place, the pattern is that AI-as-byline preprints either get moderated, get corrected on submission to a refereed venue, or remain as small-N outliers. Section 4 — AI-company self-authorship practice =============================================== When AI companies (Anthropic, OpenAI, Google DeepMind, Meta AI) publish papers and technical reports about their own models, do they list the model as an author? **The empirical answer: No, never. They list humans only.** The model itself is the *subject* of the paper, not an author of it. 4.1 — OpenAI / GPT-4 Technical Report (2023) -------------------------------------------- :Citation: OpenAI (2023). GPT-4 Technical Report. arXiv:2303.08774. :Authorship form: The paper is attributed to "OpenAI" as the corporate author. A contribution list at the end credits 279 individual human contributors (e.g. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, ...) ``[COUNT NEEDS VERIFICATION]``. GPT-4 itself is **not** listed as an author. :Implication: OpenAI describes GPT-4 throughout the paper as the *system under test*. Authorship is corporate-plus-individuals; not assigned to the model. 4.2 — Google DeepMind / Gemini (2023m12) ---------------------------------------- :Citation: Gemini Team, Google (2023). Gemini: A Family of Highly Capable Multimodal Models. arXiv:2312.11805. :Authorship form: Authored by "Gemini Team, Google" with approximately 1,350 named individual human contributors. ``[COUNT NEEDS VERIFICATION]``. Gemini itself is **not** an author. :Implication: Same pattern as OpenAI — corporate-plus-individuals, the model is the subject, not an author. 4.3 — Google DeepMind / AlphaFold (2021) ---------------------------------------- :Citation: Jumper, J., Evans, R., Pritzel, A., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. :Authorship form: Individual human authors led by John Jumper, with Demis Hassabis as corresponding author. AlphaFold itself is **not** an author. :Implication: Despite this being the canonical example of AI-accelerated biological discovery (later contributing to a Nobel Prize in Chemistry shared by Hassabis and Jumper), AlphaFold is the subject, not an author. 4.4 — Meta AI / LLaMA (2023m02) ------------------------------- :Citation: Touvron, H., Lavril, T., Izacard, G., et al. (2023). LLaMA: Open and Efficient Foundation Language Models. arXiv:2302.13971. :Authorship form: Individual human authors, led by Hugo Touvron. LLaMA itself is **not** an author. :Implication: Same pattern. 4.5 — Anthropic / Claude system cards (2024–2026) ------------------------------------------------- :Series: Claude 3 Model Card (2024m03); Claude 3.5 Sonnet Addendum (2024m06); Claude Opus 4 & Sonnet 4 System Card (2025m05); Claude Opus 4.6 System Card (2026m02); Claude Sonnet 4.6 System Card (2026); plus addenda for variants. :Authorship form: Authored by "Anthropic" (corporate) with individual contributors. Claude itself is **not** listed as an author of any of these documents. ``[CONTRIBUTOR LISTS NEED VERIFICATION for each individual document]``. :Note: Anthropic's published research papers (e.g. constitutional AI, scaling laws, mechanistic interpretability) are similarly authored by human individuals plus the Anthropic team. Claude has not been listed on a byline in any Anthropic publication located in this search. 4.6 — Summary of Section 4 -------------------------- **No AI company lists its own AI on the byline of its own papers.** This is striking because if any organisation had a strong incentive to recognise an AI as an author, it would be the labs that build them — and their commercial framing already treats their models as capable agents. Yet OpenAI, Google DeepMind, Meta AI, and Anthropic all uniformly treat the model as the *subject of study* and assign authorship to the human research team. This is consistent with the COPE / ICMJE / Nature / Science / JAMA position that the byline denotes accountability, which cannot be assigned to a model. Section 5 — Any 2024–2026 policy revisions accepting AI as author ================================================================= Has any major or specialty journal moved to *allow* AI as author since the 2023 ban-wave? **The empirical answer: No major or specialty journal has been located that allows AI as author. The post-2023 trajectory is consolidation of the no-AI-author rule, not its revision.** 5.1 — ICMJE (May 2023, January 2024 updates) -------------------------------------------- :Operative position: "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 these responsibilities are required for authorship." ``[QUOTE NEEDS VERIFICATION]`` :Disclosure rule: Authors must disclose AI use in cover letter, acknowledgements, or methods. 5.2 — COPE position on Authorship and AI tools ---------------------------------------------- :Effective date: First issued 2023m02d13; subsequent updates through 2024–2025. :Operative position: "AI tools cannot be listed as an author of a paper. AI tools cannot meet the requirements for authorship as they cannot take responsibility for the submitted work. As non-legal entities, they cannot assert the presence or absence of conflicts of interest nor manage copyright and license agreements." ``[QUOTE NEEDS VERIFICATION]`` 5.3 — Nature Portfolio ---------------------- :Operative position (2023m01, reaffirmed through 2026): "Large Language Models (LLMs), such as ChatGPT, do not currently satisfy our authorship criteria. Notably an attribution of authorship carries with it accountability for the work, which cannot be effectively applied to LLMs." ``[QUOTE NEEDS VERIFICATION]`` 5.4 — Science journals (AAAS) — Thorp editorial ----------------------------------------------- :Reference: Thorp, H. H. (2023). ChatGPT is fun, but not an author. Science 379, 313. :Operative position: "An AI program cannot be an author of a Science journal paper. A violation of this policy constitutes scientific misconduct." ``[QUOTE NEEDS VERIFICATION]`` :Subsequent update: Science later allowed *use* of AI tools as long as disclosed in methods, but did **not** change the no-AI-author rule. 5.5 — Springer Nature --------------------- :Operative position: LLMs do not satisfy authorship criteria; AI use must be documented in methods (with carve-out for AI-assisted copy-editing). 5.6 — Elsevier -------------- :Operative position: AI cannot be listed as author. AI use must be disclosed in a dedicated declaration. Elsevier's first corrigendum to remove ChatGPT from a byline (the O'Connor case in Nurse Education in Practice, Section 1.2) set the precedent. 5.7 — JAMA Network ------------------ :Operative position (2023m, updated 2024): "Nonhuman artificial intelligence, language models, machine learning, or similar technologies do not qualify for authorship." ``[QUOTE NEEDS VERIFICATION]`` Applies across all 14 JAMA Network titles. 5.8 — The Lancet / Lancet Digital Health ----------------------------------------- :Operative position: "AI tools cannot be listed as authors, as authorship implies responsibilities and tasks that can only be attributed to and performed by humans." ``[QUOTE NEEDS VERIFICATION]`` Lancet Digital Health is the most AI-relevant specialty title within the Lancet family and still applies this rule. 5.9 — NEJM AI (launched 2024) ----------------------------- :Launch date: 2024 :Operative position (editorial policies): "Because authors are responsible for accuracy, integrity, and originality of the work, chatbots or other AI-assisted technologies cannot be listed as authors. Authors should carefully review and edit all materials produced through the use of AI, to prevent the submission of authoritative-sounding output that is incorrect, incomplete, or biased." ``[QUOTE NEEDS VERIFICATION]`` :Distinctive editorial stance: Unlike the parent NEJM, NEJM AI actively **encourages** authors to use LLMs in manuscript preparation (see the editorial "Why We Support and Encourage the Use of Large Language Models in NEJM AI Submissions", DOI 10.1056/AIe2300128 ``[DOI NEEDS VERIFICATION]``). But this encouragement is for *tool use with disclosure*, **not** for AI on the byline. NEJM AI's authorship rule explicitly forbids AI as author. **Key takeaway about NEJM AI.** A journal that was specifically launched to publish AI-relevant work, by a publisher willing to break from its parent title's traditional caution, *still* does not allow AI on the byline. This is significant because NEJM AI is the most plausible candidate for a journal that might have revised the consensus — and it did not. 5.10 — Summary of Section 5 --------------------------- Across every major umbrella body (ICMJE, COPE), every major publisher (Elsevier, Springer Nature, Wiley implicit via society journals), and every flagship general journal (Nature, Science, JAMA, BMJ, Lancet, NEJM), as well as the newest AI-specialty journal (NEJM AI, launched 2024), **the no-AI-author rule is in force and has not been revised**. The 2023–2026 trajectory is one of **consolidation**, not relaxation. Section 6 — Calls to revise the consensus ========================================= Has any prominent voice in the academic literature (2023–2026) argued for revising the no-AI-author rule to allow AI on the byline? 6.1 — Hosseini, Gordijn, Kaebnick & Holmes (2025) -------------------------------------------------- :Citation: Hosseini, M., Gordijn, B., Kaebnick, G. E., & Holmes, K. (2025). Disclosing generative AI use for writing assistance should be voluntary. Research Ethics. DOI 10.1177/17470161251345499 ``[DOI NEEDS VERIFICATION]``. :Position: This paper argues that **disclosure** of generative AI use for *writing assistance* should be **voluntary**, not mandatory. It is a departure from these authors' earlier (2023) stance that disclosure should be mandatory. :Does it argue for AI authorship? **No.** The paper is about disclosure thresholds, not about authorship attribution. The authors explicitly do not argue that AI should be on the byline. This is the most prominent recent dissent from the disclosure norm, and it does not call for the no-AI-author rule to be revised. 6.2 — "Responsibility is not required for authorship" (PMC 12015057) -------------------------------------------------------------------- :Reference: A 2024–2025 commentary that challenges the COPE / ICMJE premise that *accountability* is a necessary criterion for authorship. ``[FULL CITATION NEEDS VERIFICATION]``. :Position: Argues that the criterion COPE / ICMJE use to exclude AI (i.e. that AI cannot be accountable) is too strong; many human authors do not in practice satisfy strict accountability either. :Does it argue for AI authorship? **Indirectly** — it weakens the exclusion argument but stops short of directly advocating AI on the byline. 6.3 — Other voices (2023–2026) ------------------------------ Multiple commentaries and editorials in this period (in journals including Accountability in Research, Learned Publishing, Research Ethics, ESMO Open, AMEE Guide series, Journal of Scholarly Publishing, Behavioral Ecology and Sociobiology, Radiology, CMAJ) discuss the AI-authorship question. The consistent pattern is: - Most argue for *strict no-AI-author* with mandatory disclosure. - A minority argue for the *Hosseini-style voluntary disclosure*. - A very small minority (essentially the original Zhavoronkov line, occasionally echoed) treats the question as legitimately open. **No prominent academic voice located in 2023–2026 makes a sustained, formal argument that AI models should be listed on the byline of refereed papers as currently composed**, with the operational details of how accountability transfer would work. The closest is the "responsibility is not required for authorship" line of argument, which is a *negative* argument (against the exclusion criterion) rather than a *positive* proposal (for inclusion). 6.4 — Summary of Section 6 -------------------------- The academic literature contains active debate about *how much* disclosure is required and about *whether the accountability criterion is strictly correct*. It does **not** contain a sustained prominent argument *for* allowing AI on the byline. The published calls to revise the consensus are at most calls to lower the disclosure bar, not to admit AI as author. Section 7 — Bottom-line answer ============================== The lead question for this fact-sheet was: *"I'd like to know if I'm really the first one to push this or e.g. there are already a ton of AI co-authorship papers e.g. on arxiv or biorxiv or somewhere."* **Answer: A very small number of precedents exist (almost all from a narrow late-2022 to early-2023 window before the policy wave); most were corrected; only one durable instance has been located; and no refereed journal has accepted an AI on the byline since the May 2023 ICMJE update.** More precisely, in the three-tier classification: - **"No precedent located" — false.** A handful of precedents do exist. See Section 1. - **"A few precedents exist but were retracted or corrected" — partially true.** Of the three confirmed 2022m12–2023m01 cases: - Kung et al. (medRxiv v1, 2022m12d21): ChatGPT byline dropped between preprint and journal version. **Corrected.** - O'Connor et al. (Nurse Education in Practice, 2023m01): ChatGPT byline removed by Elsevier corrigendum. **Corrected.** - Zhavoronkov & ChatGPT (Oncoscience, 2022m12d21): ChatGPT byline appears to remain in PMC and journal as of 2026m05d13. **Not corrected**, but cited in the literature as a cautionary example. - **"Several precedents exist in venue X, Y, Z" — false.** No durable cluster of refereed precedents exists. The one durable case (Oncoscience) is widely treated as the outlier that *triggered the ban*, not as a precedent supporting the practice. **Therefore, the most accurate statement is:** *Officially-acknowledged AI co-authorship at a refereed venue is essentially unprecedented as of 2026m05d13. One marginal precedent (Oncoscience 2022m12d21, Zhavoronkov & ChatGPT) exists but is widely regarded as the case that crystallised the no-AI-author consensus rather than as a precedent legitimising the practice. All other early-2023 attempts were either corrected before journal acceptance or corrected after via corrigendum. No major or specialty refereed journal — including the newly-launched AI-specialty title NEJM AI — currently permits AI on the byline. No prominent academic voice has published a sustained formal argument for revising this consensus to allow AI authorship. AI companies (OpenAI, Google DeepMind, Meta AI, Anthropic) do not list their own models as authors of their own papers.* **Implication for the b19 discussion.** Proposing officially-acknowledged AI co-authorship of a refereed paper in 2026m05 is **substantively unprecedented**. It is not "everyone is doing this." It is not "there is a building wave." The opposite is true: the post-2023 trajectory has been consolidation of the no-AI-author rule across all major publishers, all major umbrella bodies, and all major specialty venues — including the newest AI-dedicated specialty title — with no notable counter-trend. The investigator who seriously pushes for AI co-authorship at a refereed venue would, to the best of this fact-sheet's reconnaissance, be among the first to do so as a deliberate, framework-grounded proposal (as opposed to the 2022m12–2023m01 ad-hoc experiments that triggered the bans). Verification flags summary ========================== The following items were noted in the fact-sheet as needing direct verification (because WebFetch was unavailable in this subagent session and policy / DOI / quote details were summarised from WebSearch snippets): - ``[DOI NEEDS VERIFICATION]``: 10.1101/2022.12.19.22283643 (Kung et al. medRxiv); 10.1371/journal.pdig.0000198 (Kung et al. PLOS Digital Health); 10.1016/j.nepr.2022.103537 (O'Connor); Oncoscience DOI for Zhavoronkov. - ``[DATE NEEDS VERIFICATION]``: Nurse Education in Practice issue/date for O'Connor article. - ``[QUOTE NEEDS VERIFICATION]``: COPE position quote; Nature policy quote; Science / Thorp quote; JAMA quote; Lancet quote; NEJM AI editorial quote; arXiv blog post quote; Elsevier corrigendum quote. - ``[COUNT NEEDS VERIFICATION]``: 279 contributors on GPT-4 Technical Report; ~1,350 contributors on Gemini paper. - ``[CONTRIBUTOR LISTS NEED VERIFICATION]``: Anthropic Claude system card contributor lists. - ``[CASE NEEDS VERIFICATION]``: "Iskender SSRN" reference; fourth case in Nature's "at least four" count. - ``[FULL CITATION NEEDS VERIFICATION]``: "Responsibility is not required for authorship" commentary (PMC 12015057). - ``[POLICY TEXT NEEDS VERIFICATION]``: ChemRxiv policy detail. None of these flags affect the bottom-line answer in Section 7, which rests on the convergent evidence pattern rather than on any single quoted item. Sources consulted (WebSearch) ============================= Search snippet returns from the following sources were used to compose this fact-sheet. Original direct fetching (WebFetch) was not available; therefore quotations are paraphrased and flagged above. - Nature: Stokel-Walker, C. (2023). ChatGPT listed as author on research papers: many scientists disapprove. Nature 613, 620–621. - Nature Portfolio editorial policies on AI. - Science: Thorp, H. H. (2023). ChatGPT is fun, but not an author. Science 379, 313. DOI 10.1126/science.adg7879. - ICMJE: Recommendations updates 2023m05 and 2024m01 (icmje.org). - COPE: Position statement "Authorship and AI tools" (publicationethics.org), first issued 2023m02d13. - JAMA Network: Guidance for Authors, Peer Reviewers, and Editors on Use of AI (jamanetwork.com, 2023; updates 2024). - The Lancet editorial policies (thelancet.com). - NEJM AI editorial policies (ai.nejm.org/about/editorial-policies); editorial "Why We Support and Encourage the Use of Large Language Models in NEJM AI Submissions" DOI 10.1056/AIe2300128 ``[DOI NEEDS VERIFICATION]``. - arXiv blog: "arXiv announces new policy on ChatGPT and similar tools" 2023m01d31 (blog.arxiv.org). - bioRxiv FAQ. - Elsevier generative AI policies. - Springer Nature AI guidance for researchers (group.springernature.com). - Kung et al. (2023). Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. PLOS Digital Health 2(2): e0000198. - Kung et al. medRxiv preprint v1, 2022m12d21. - O'Connor, S. & ChatGPT (2023). Open Artificial Intelligence Platforms in Nursing Education: Tools for Academic Progress or Abuse? Nurse Education in Practice, Vol. 66. - Zhavoronkov, A. & ChatGPT (2022). Rapamycin in the context of Pascal's Wager. Oncoscience 9, 2022m12d21. PMC9796173. - Thunström, A. O., Steingrimsson, S., & GPT-3 (2022). Can GPT-3 write an academic paper on itself, with minimal human input? HAL preprint hal-03701250. - OpenAI (2023). GPT-4 Technical Report. arXiv:2303.08774. - Gemini Team, Google (2023). Gemini: A Family of Highly Capable Multimodal Models. arXiv:2312.11805. - Jumper, J., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. - Touvron, H., et al. (2023). LLaMA: Open and Efficient Foundation Language Models. arXiv:2302.13971. - Anthropic Claude system cards (anthropic.com), 2024–2026. - Hosseini, M., Gordijn, B., Kaebnick, G. E., & Holmes, K. (2025). Disclosing generative AI use for writing assistance should be voluntary. Research Ethics. DOI 10.1177/17470161251345499 ``[DOI NEEDS VERIFICATION]``. - "Responsibility is not required for authorship." PMC12015057. ``[FULL CITATION NEEDS VERIFICATION]``. - Retraction Watch coverage of AI-related corrections (retractionwatch.com). - Sallam, M. (2023). ChatGPT Utility in Healthcare Education, Research, and Practice. Healthcare 11, 887. (Author of a meta-analysis; does *not* list ChatGPT as author.) - Defining the Boundaries of AI Use in Scientific Writing (PMC 12170296, 2025 review of editorial policies). End of Fact-sheet 6 ===================