Nature, Science, or Cell allows AI systems as authors

Where things stand (checked 16 Oct 2026)

All three journals ban AI authors. Each restated the ban in 2026, most recently around the question's opening date.

• Nature / Springer Nature. A new "traffic-light" AI policy came out on 10–11 Sep 2026. It puts "listing AI as an author" and "assigning authorship or accountability to AI systems or tools" in the Red – not permitted category (Nature Portfolio AI policy; Springer Nature, AI use in manuscript preparation). Springer Nature's head of journals policy wrote that responsibility "cannot be delegated to AI systems, regardless of how sophisticated those systems may become" (Times Higher Education, Sep 2026).
• Science. The editorial-policies page, in a Wayback Machine copy from 14 Jun 2026, says "artificial intelligence tools cannot be authors" and that AI tools "may not be listed as authors or coauthors." Editor-in-chief Holden Thorp's editorial of 16 Jul 2026 ("AI in scientific publishing: Slower, worse, and more expensive", doi:10.1126/science.aek5570) is strongly sceptical of AI.
• Cell / Elsevier. Elsevier's generative-AI policy, marked "updated June 2026", says authors "should not list AI tools as an author or co-author." Cell Press policy pages say AI "must not" be listed as an author.
• The editors in person. On 22 Sep 2026, at a Broad Institute panel (YouTube Q04PAohvSYU, transcript pulled), all three editors-in-chief said the same thing:
  • John Pham (Cell): "AI agents cannot be authors on papers because they can't hold them responsible."
  • Magdalena Skipper (Nature): credit and accountability are two sides of one coin, "that's why an agent cannot be credited."
  • Thorp (Science): asking about credit is "incomplete" if you cannot assign responsibility.

So the near term is close to zero. These are rules based on principle, restated within the last month.

Base rate

From January 2023 to October 2026, none of the three, and no major publisher, has allowed AI authorship, across several policy revisions. The Committee on Publication Ethics (COPE) and the medical editors' committee (ICMJE, updated January 2026) both still reject it. The only venues that accept AI authors are fringe ones: the Agents4Science conference and the aiXiv preprint server. The recent trend is toward tighter rules: Philosophy & Public Affairs banned AI-authored content in August 2026. US law also requires a human author for copyright; the Supreme Court declined to hear Thaler v. Perlmutter in March 2026. Core authorship rules usually change over decades, not years.

Paths to YES and NO

YES. AI capability is rising very fast in 2026:

• OpenAI says its agents solved the Navier–Stokes Millennium Prize problem (8 Sep), and it then released 722 AI-generated maths papers.
• Anthropic says its AI found a CRISPR-like enzyme system.
• Amodei writes that AI is beginning to improve itself.

If AI produces most frontier science by 2029–31, calling only humans the authors could start to look dishonest. New editors, or competition among journals for papers from AI labs, could then lead one journal to create an AI-author category with a human guarantor. That would count under the resolution rules.

NO. The objection is about accountability, not capability. Editors can, and have said they will, handle AI work through disclosure, contribution statements and human guarantors; none of these count. The current mood (AI-generated junk papers, fabricated citations, the OpenAI–Hugging Face agent incident, calls to slow AI down) pushes against giving AI any standing.

Estimate

Rough scenario mix:

• AI becomes the main engine of research by 2029–30 (about 40%), with about an 11% chance of a qualifying policy in that world.
• Paced progress (40%), about 2%.
• Backlash or restriction (20%), about 0.5%.

That gives roughly 5%.

Quarterly hazards, computed in Python:

• 0.08% for Q4 2026
• rising to about 0.2% per quarter in 2027
• about 0.3% per quarter from 2028 to 2031

Cumulative probability: about 0.8% by end-2027, 1.9% by end-2028, 4.4% by end-2030 and 5.7% by end-2031.

Strongest case against this estimate: progress is unusually fast, there are three journals, and the bar includes conditional or pilot policies. If AI clearly out-researches humans by about 2029, attribution norms could flip faster than history suggests. I allowed for this with hazards that rise over time instead of a flat near-zero rate.

Question. By the last day of each quarter, has Nature, Science, or Cell published a change to its authorship policy allowing an AI system to be listed as an author of a research paper?

Definitions.

• Nature, Science, or Cell: the flagship journals with those exact titles, not other Nature Portfolio, Science-family, or Cell Press journals. A publisher-wide policy counts only if it applies to at least one of these three journals.
• AI system: a computational artificial-intelligence model, agent, or system, from any developer, identified as such rather than as a human, institution, or human research consortium.
• Listed as an author: credited as an author or coauthor in the paper's formal author list, including an explicitly designated AI-author category within that list. Acknowledgments, contribution statements, methods disclosures, software citations, and separate lists of tools do not constitute authorship.
• Research paper: a peer-reviewed paper reporting original research, including original theoretical or computational research. Reviews, editorials, news, opinion pieces, fiction, and demonstrations published solely to discuss AI authorship do not count.
• Policy change: an adopted revision that expressly permits AI authorship, either generally or under stated conditions, for research papers in a qualifying journal. Conditional permission, including requirements for human coauthors or human acceptance of responsibility, counts. A final adopted policy with a specified future effective date counts on publication; proposals and consultations do not.

Resolves YES if a qualifying journal or its publisher publishes an official authorship policy, author instructions, or authoritative editorial announcement establishing such a policy change. Permission under an explicit exception or pilot policy counts if it authorizes AI systems to appear as authors of original research papers. No actual paper with an AI author need have been published, and no AI system need yet have satisfied any eligibility conditions.

Resolves NO if no qualifying change has been published by the quarter-end. In particular, the following do not qualify:

• Permission to use AI in research, writing, reviewing, or image generation without permission for formal AI authorship.
• Removal of an explicit AI prohibition without an affirmative policy statement permitting AI authorship.
• Calls for policy reform, predictions about future eligibility, or statements that the journal is considering a change.
• An AI name appearing on a paper through error, undisclosed use of a pseudonym, or an isolated editorial waiver without an official statement establishing a changed policy.
• Policy changes restricted to sister journals or to nonresearch content.

Resolving source. Official publications of the journal or its publisher only: Nature/Springer Nature's Nature Portfolio Artificial Intelligence policy and authorship policies; AAAS's Science Journals: Editorial Policies; and Cell Press's Cell Journal Policies, including applicable Elsevier Generative AI policies for journals. Official replacement pages, author instructions, and editorials explicitly announcing adopted policy changes also qualify. This is a narrower source requirement than Public confirmation (A2). Apply Resolution date (A3) and the Resolution window. For an undated policy-page revision, a dated official announcement or publicly timestamped archived copy must establish the revised wording; use the earliest such documented public date, without inferring an earlier change date.