CoChat
Sign Up
Cited by 15Academic Publishing and Open AccessEthics and Social Impacts of AIAcademic Writing and Publishing

Resisting AI slop

Open access
Thorp, H. Holden·2026·Science
DOI 10.1126/science.aee8267
Open this paper in CoChat — free accountOpen PDF

Abstract

It’s hard to talk about any topic in science or education today without the subject of artificial intelligence (AI) coming up—whether large language models should be allowed to aid in searching for a scientific paper or even to write or review the paper itself. In some of the wildest speculations, the humans involved in conducting scientific studies and experiments and vetting the results for publication will be steadily eliminated from the process. But when such grandiose rhetoric starts flying, we at Science try to keep calm and carry on in contributing to a robust, human-curated research literature that will stand the test of time.

AI snapshot

Written from the paper’s full text.

TL;DR

Thorp, Editor-in-Chief of the Science journals, argues that AI can help scientific publishing if used carefully, but that human judgment and responsibility remain essential to a trustworthy literature. He outlines Science's current AI policies and reports early data on data-sharing compliance.

Key findings

  • Science permits undisclosed use of large language models for editing text for clarity and readability and for gathering references, but use beyond that (e.g., drafting manuscript text) must be declared; AI-generated figures are not allowed, and all authors must certify responsibility for all content, including AI-assisted content.
  • In a collaboration with DataSeer evaluating adherence to Science's data- and code-sharing mandate, 69% of 2680 Science papers published between 2021 and 2024 shared data.
  • A DataSeer reproducibility checklist, tested in a 2025 pilot, is being integrated into journal protocols; DataSeer's natural language processing scans a paper and generates a prefilled checklist that authors confirm and revise.
  • Science prohibits reviewers from putting any part of a paper into a large language model to generate a review, though reviewers may use AI to refine text if declared; AI tools such as iThenticate and Proofig are used to detect plagiarism and altered figures, but their output "require[s] more human effort, not less."

Methods

  • This is an editorial/commentary by the Editor-in-Chief, not an empirical study; it describes journal policy and reports descriptive compliance figures.
  • The only quantitative data reported are from a DataSeer evaluation of data sharing across 2680 Science papers published 2021–2024 (69% shared data) and a 2025 pilot of the DataSeer reproducibility checklist.
  • No formal analysis, sample selection, or statistical method is described.

Limitations

  • Stated by the paper: "no system, human or artificial, can catch everything," so degradation of the literature cannot be fully prevented.
  • Evident from the text: the piece is an opinion editorial rather than a study, so its claims about AI's effects on publishing and employment are framed as skepticism and prediction ("I remain skeptical," "Acceptance of bombastic statements... should wait for verification") rather than demonstrated findings; the 69% data-sharing figure is reported without detail on how compliance was measured.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1126/science.aee8267

Cites

Statistics

Nothing to recompute.

Claims the paper makes

  • Findingown data

    Of 2680 Science papers published between 2021 and 2024, 69% shared data.

    “The initial results are encouraging in that of 2680 Science papers published between 2021 and 2024, 69% shared data.”
    Quote in the paperNot a substantial claim
  • Study design and conductreasoning

    A DataSeer reproducibility checklist, tested in a 2025 pilot program, is being integrated into Science's protocols.

    “To further improve transparency, a DataSeer reproducibility checklist, which Science tested in a 2025 pilot program, is being integrated into the journal's protocols.”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. 13
  • May combine two claims. 1

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingnone given

    AI is helping Science catch errors that can be corrected or elements that are missing from a paper but should be included, such as supporting code or raw data.

    “Although AI is helping Science catch errors that can be corrected or elements that are missing from a paper but should be included, such as supporting code or raw data”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingreasoning

    AI is allowing Science to identify problems more rigorously than before.

    “AI is allowing Science to identify problems more rigorously than before”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Science allows reviewers to use AI to refine text, as long as the use is declared.

    “However, Science does allow reviewers to use AI to refine text, as long as the use is declared.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductnone given

    Science prohibits reviewers from putting any part of a paper into a large language model to generate a review.

    “Unlike some new efforts, Science prohibits reviewers from putting any part of a paper into a large language model to generate a review.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The use of AI and the evaluation of its output require more human effort, not less.

    “its use and the evaluation of the output require more human effort, not less.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Science's policies require declaration for use of AI beyond the processes allowed without disclosure, for example in drafting manuscript text.

    “However, the use of AI beyond that-for example, in drafting manuscript textmust be declared.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    A bigger concern about AI is that its use in the production or review of research papers could degrade the reliability of the scientific literature.

    “A bigger concern about AI is that its use in the production or review of research papers could degrade reliability of the scientific literature.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Science's most recent policies allow the use of large language models for certain processes without any disclosure, such as editing the text in research papers to improve clarity and readability or assisting in the gathering of references.

    “Science's most recent policies allow the use of large language models for certain processes without any disclosure, such as editing the text in research papers to improve clarity and readability or assisting in the gathering of references.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Science requires all authors to certify and be responsible for all content, including content generated with the aid of AI.

    “All authors must certify and be responsible for all content, including that generated with the aid of AI.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    As a small family of journals that can put more human effort into each paper, the Science journals are less susceptible, and contribute less, to the accumulation of AI slop in the literature.

    “As a small family of journals that can put more human effort into each paper, the Science journals are less susceptible-and contribute less-to the accumulation of”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: No system, human or artificial, can catch everything.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Science's policies do not allow the use of AI to create figures.

    “And the use of AI to create figures is not allowed.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Over the past year, Science collaborated with DataSeer to evaluate adherence to its policy mandating the sharing of underlying data and code for all published research articles.

    “Over the past year, Science has collaborated with DataSeer to evaluate adherence to its policy mandating the sharing of underlying data and code for all published research articles.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Science uses AI tools, such as iThenticate and Proofig, to better identify plagiarized text or altered figures.

    “Science also uses AI tools, such as iThenticate and Proofig, to better identify text that has been plagiarized or figures that have been altered.”
    Quote in the paperNot a substantial claim
13 of 13

  • The paper relies on this without support.

    Most technological advances have not led to catastrophic job losses.

    “Most technological advances have not led to catastrophic job losses.”

  • The paper relies on this without support.

    No system, human or artificial, can catch everything.

    “no system, human or artificial, can catch everything.”

  • The paper relies on this without support.

    Online courses became an important element of education and allowed universities to grow, not shrink.

    “online courses did become an important element of education and allowed universities to grow, not shrink.”

  • The paper relies on this without support.

    The movement of journals to publishing online increased the size and scale of scholarly publishing.

    “The movement of journals to publishing online provoked a similar result-it increased the size and scale of scholarly publishing.”
4 of 4

Cited by

Sorted by citations

Related

Authors
Thorp, H. H.
Affiliations
H. Holden Thorp is Editor-in-Chief of the Science journals.
Corresponding author
Thorp, H. Holden
ORCID
Thorp, H. Holden
Venue
Science
Publisher
American Association for the Advancement of Science (AAAS)
Published
2026-01
Volume
391
Issue
6780
Pages
5-5
ISSN
  • 0036-8075 (ISSN-L)
  • 1095-9203
DOI
10.1126/science.aee8267
OpenAlex
W7117884547
Cited by
14
References
0
Citations by year
2026: 14
Type
journal article
Language
en
Open access
Locations
  • Science (publishedVersion)
  • PubMed (publishedVersion)
Topics
  • Social Sciences › Decision Sciences › Information Systems and Management › Academic Publishing and Open Access
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Arts and Humanities › History and Philosophy of Science › Academic Writing and Publishing
Keywords
artificial intelligence in science, large language models, scholarly publishing, peer review, AI-generated research

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-29.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-29.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review