Can AI Be an Author? Math Paper Credit Rules (2026)

Quick answer: No. Under the major publishing policies in force in November 2026, AI systems such as ChatGPT or Claude cannot be listed as authors. arXiv, the Committee on Publication Ethics (COPE), Springer Nature and the ICMJE all require accountable human authors and transparent reporting of significant AI use. If an AI produced a proof, the paper can say so plainly, as several 2026 papers did (“The result was proved by GPT-6 Astra”), but the human authors remain responsible for its correctness. To acknowledge AI, name the system, say exactly what it did, explain how its output was checked, and follow the venue’s rules.

Facts checked November 22, 2026.

In 2026, AI systems stopped being only calculators and started supplying key steps, and sometimes whole proofs, in real research papers. That raises a question mathematics has never had to answer: if a machine proves a theorem, who gets the credit? Here are the current rules, real 2026 examples, the arguments on each side, and a simple tool for writing an honest AI disclosure.

Authorship, Contribution and Acknowledgment Are Different

Much confusion disappears once you separate three ideas. Authorship is a formal role with legal and ethical duties: approving the final paper, declaring conflicts of interest and correcting errors. Contribution is who actually supplied ideas, computations or proofs. Acknowledgment is how a paper records help that does not amount to authorship. Current rules say an AI can make a large contribution, which must be acknowledged, but cannot hold authorship.

The Rules: AI Is Credited, Not Listed as an Author

OrganizationCan AI be an author?What it requires
arXivNoReport significant use of generative AI; every author takes “full responsibility for all its contents, irrespective of how the contents were generated”
COPE (Committee on Publication Ethics)NoAI tools cannot take responsibility, so they cannot meet authorship criteria; disclose use
Springer Nature and many journalsNoDocument AI use in the methods or acknowledgments
ICMJE (medical journal editors, widely cited)NoChatbots cannot be authors; disclose writing help in acknowledgments and research use in methods

The logic is the same everywhere: AI systems cannot take on the legal, ethical and scholarly obligations of authorship, including final approval, conflict-of-interest declarations and responsibility for correcting errors. So an AI gets credit the way a crucial instrument or software package does, by being named and described. Specific disclosure requirements differ, so always check your target journal.

Real Cases From 2026

Paper (arXiv)DateHow the AI is credited
Mubayi and Verstraete, “Erdős-Sós for digraphs”September 10, 2026Abstract: “The result was proved by GPT-6 Astra.” Two human authors
James Leng, new proof of two-color regularity of Pythagorean triplesSeptember 28, 2026Abstract: “The proof was found by ChatGPT Astra.” One human author
Edward Witten, lattice definition of chiral gauge theoriesNovember 3, 2026In the text: “Anthropic’s Claude Opus 5.5 suggested a likely variant”; acknowledgments: “Assistance on many details was provided by Claude Opus 5.5”

In all three, the humans are the authors and the AI is credited openly in the paper itself. The difference is in how much the AI did: from “assistance on many details” to “proved the result.” Note what each claim covers: Mubayi and Verstraete credit the AI with proving a new theorem, while Leng’s paper gives a new proof of a theorem already known since 2016. For that story, see the 200-terabyte math proof.

What Does a Human Author Contribute When AI Proves It?

Even when a model writes the proof, someone has to:

  • Choose the question and see that it matters.
  • Frame it precisely, so the AI is proving the right statement.
  • Check the proof, by hand or in a proof assistant like Lean, and take responsibility if it is wrong.
  • Place it in context: what is new, how it connects to known results, and what follows.

Formal verification can help: a Lean-checked proof shows correctness, but it does not decide who contributed what. That is why rules focus on responsibility rather than on who typed which line. An unchecked AI proof posted under a human name is the real risk, and a September 2026 audit of OpenAI’s AI-generated results relied on independent human referees for exactly this reason. See can we trust AI math proofs?

How to Acknowledge AI in a Math Paper

A good AI-use statement answers five questions:

  1. Which system? Provider, model name and version.
  2. What did it do? Language editing, literature search, computations, ideas, proof steps or the main proof.
  3. How much? Did it assist human work, or produce a major result?
  4. How was it checked? Independent human checking, code runs, or a formal proof in Lean.
  5. Who is accountable? State that the human authors take responsibility.
Type of AI helpWhere to disclose it
Wording, grammar, LaTeXAcknowledgments or the journal’s AI-use declaration
Computations, code, literature searchMethods section or where the computation is used
Key lemma or ideaIn the text at that step, plus acknowledgments
Main proof or resultAbstract or introduction, as well as the body

Write an Honest AI Disclosure

Choose what an AI system did in your work, and get a suggested disclosure sentence and a credit category.



What did it do?

This is a writing aid, not an official policy; always follow the rules of the journal or archive you submit to.

The Debate

The case for crediting humans as authors

  • Responsibility: only people can be accountable for errors.
  • Mathematicians have always used tools, from tables to computer algebra, without listing them as authors.
  • Choosing and verifying the problem is real intellectual work.

The case for rethinking credit

  • If an AI “proved the result,” calling a human the sole author can mislead hiring and grant committees about who did what.
  • Careers are built on authorship, so unclear credit may reward whoever has the best AI access.
  • Some propose new labels, such as separate categories for human, AI-assisted and AI-generated work, or “co-ownership” of ideas instead of authorship.

Some go further. In “The Crisis of AI-Generated Mathematics” (arXiv, August 2026; revised in October to the version accepted by the Notices of the American Mathematical Society), Max Weinreich makes “the case for total opposition to the use of artificial intelligence in mathematics,” arguing that AI tools “short-circuit human understanding.” That is one individual’s position, not a consensus. Meanwhile, new repositories such as Hexagon, launched in fall 2026, record where research works came from, including AI-assisted ones. The community has not settled how credit should work.

What It Means for Students

  • Homework and contests: follow your teacher’s or contest’s rules; unacknowledged AI help is usually treated like copying. See is using AI for math homework cheating?
  • Research projects: keep notes on what the AI did and disclose it.
  • Understanding matters most: you are responsible for every step you submit, so make sure you can explain it.

Common Misconceptions

“Using AI makes a paper invalid.” No. Disclosed AI use is allowed on arXiv and by most publishers; the authors must check the work.

“If AI did most of the work, it should be an author.” Not under current rules, which tie authorship to accountability. Credit goes in the text and acknowledgments.

“Acknowledging AI is optional.” For significant use, arXiv and most journals require it.

Watch: Terence Tao on AI and Credit

If you learn better by watching, these videos are worth your time.

No More Heroes – Terence Tao Talks to Numberphile About AI (Numberphile). Terence Tao on how AI is changing who does mathematics and how credit works.
Terence Tao on how AI math solutions miss the point (SciFri). Why a correct AI proof is not the same as human understanding.

Check Your Understanding

Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.

Answer each question and press Submit to see your result right away.

1. Can an AI system be listed as an author on arXiv?
Found a bug? Let us know!
2. Why do rules tie authorship to people?
Found a bug? Let us know!
3. Which 2026 paper says “The result was proved by GPT-6 Astra”?
Found a bug? Let us know!
4. Where did Edward Witten credit Claude Opus 5.5?
Found a bug? Let us know!
5. What should a student do after using AI on a research project?
Found a bug? Let us know!

Frequently Asked Questions

Can AI be an author of a math paper?

No, under current rules. arXiv, COPE and major publishers say AI tools cannot be authors because they cannot take responsibility; significant AI use must be disclosed instead.

Who gets credit when AI proves a theorem?

The human authors who posed the question, checked the proof and take responsibility, with the AI’s role described in the paper or acknowledgments.

How should I acknowledge AI in a math paper?

Name the system, say exactly what it did (for example, suggested a lemma or produced the main proof), and state that the authors checked the work and take responsibility.

Has an AI proved a published theorem in 2026?

Several 2026 arXiv preprints credit AI with proofs, such as Mubayi and Verstraete’s Erdős-Sós for digraphs, which says the result was proved by GPT-6 Astra. These are preprints, not yet peer-reviewed journal articles.

Where should I put an AI disclosure in a paper?

Writing help usually goes in the acknowledgments or an AI-use declaration; computations in the methods; a key AI-found step in the text at that point; and an AI-generated main result in the abstract or introduction too.

Does arXiv ban AI-generated papers?

No outright ban as of November 2026. arXiv permits AI as a tool if its use is disclosed and the work meets its standards, and since October 1, 2026 limits each submitter to two new submissions a month.

Keep Exploring

Sources and Further Reading

Related to This Article

What people say about "Can AI Be an Author? Math Paper Credit Rules (2026) - Effortless Math"?

No one replied yet.

Leave a Reply