A new type of paper mill

Eric Dolores-Cuenca, Postdoc at Yonsei University

The purpose of this letter is to warn the mathematical community of a new type of paper mill that could potentially appear as a consequence of the advances in artificial intelligence.  I am a postdoc working on Math and AI, and I organize an international online seminar on math and machine learning.

Firstly, there exist groups that produce fake research papers and sell them to scientists; these groups are commonly called paper mills. According to this study, at least 2% of all research submissions in all sciences are submitted by paper mills.

Secondly, there also exist start-ups, companies and universities working on solving mathematical problems with machine learning. In particular, some use tools that take information from a mathematical problem and return, when possible, experiments to prove it. The level of human interaction varies: in some cases the user has to provide only a text description of the problem; while for other programs, the user must write code that evaluates the fitness of the solutions proposed by agents.1  AI agents are also capable of transforming the output of those experiments into a written research paper. While it is not clear how much interaction was needed to produce the preprint about the Navier-Stokes problem or to solve the Jacobian conjecture, in this post I want to give emphasis on the possibility of an artificial intelligence solving completely a problem only using a prompt as input.

With these two antecedents, I would like to ask the reader: what would happen if a group trained an algorithm to recognize important mathematical questions? The algorithm can work in parallel and mass produce meaningful questions. If we now give this list of questions as inputs to algorithms that try to solve them, and we furthermore assume that they manage to solve a small percentage of the input questions, then the final products are research papers that are publishable since they have solved meaningful questions for mathematicians. Finally, what if the company sells those finished results to mathematicians? This would create a paper mill that produces correct papers and sells the authorship. Moreover, if an AI that solves problems has a certain level of quality, an AI that formalizes the proof can further filter the papers.

One can wonder what the danger is if paper mills are producing knowledge, that is, publishable papers. Fraudulent credit is one of the problems; affecting hiring, promotion, grants, and reducing the credibility of the publication process.

In informal conversations, I asked some of the experts who attended the Mathematics × AI: Challenges and Opportunities event at the Korea Institute for Advanced Study and the International Conference on Machine Learning 2026 about the possibility of this new type of paper mill. Some of the people I talked to are collaborating with AI companies, while others were participants with different levels of experience in AI.

While I personally expected to be assured of the impossibility of this new type of paper mill, some of the experts considered this scenario plausible within a few years, and they could not think of a solution for journals to detect this problem. Some told me they have students working on an algorithm to predict important problems. I was encouraged by them to keep talking about this problem in order to raise awareness, which led me to write this letter.

Why would this be a very difficult problem for journals?

From the point of view of the journals, there are no steps in the review process to ensure that the person who submitted the paper did not buy the paper from a paper mill, or to certify that the person is aware of the content of the paper. Note also that using a detector to see if a paper is written by AI is useless: a person can buy finished papers and rewrite them.

We are now seeing talks in which people used AI in their work. Note that the quality of the talk may not be related to the use of AI. The speaker could be afraid of public speaking, perhaps the speaker does not have enough experience giving talks, faces a language barrier, or is new to the topic and knows just the minimum required for the result to hold.

I call on the community to discuss what changes journals would need to implement if we ever reach that level of technology.

Acknowledgement: I would like to thank Susana Lopez Moreno for related discussions.

  1. The following examples, listed alphabetically, are not exhaustive: Aletheia, AlphaEvolve, Astra, Iteris, OpenEvolve, Rethlas, and ShinkaEvolve. ↩︎

Received 11 September 2026.

7 responses to “A new type of paper mill”

  1. hopeful_mathematician Avatar
    hopeful_mathematician

    I think the answer is the same as the answer to most issues with journals mentioned already.

    If journals exist at all, they might publish a vanishingly small number of articles relative to the number of standard reserach artifacts made available via some other means (papers in repositories, projects in github, something new entirely). So that peer review can handle them. Hopefully we will find good ways to make “non-journal-published” artifacts available to search, etc.

    Most likely, the “journal” articles are not so important for careers but are to communicate something that should genuinely have a broad audience.

    We will find some other way to evaluate researchers based on research artifacts, perhaps taking a retrospective view and/or actually discussing in more depth with early career researchers rather than relying on a publication record. In such a landscape, anyone who wants to contribute in any way, can do so. If someone writes a program that spits out useful science, that’s fine too. I don’t think we’re going to run out of science. Is that possible?

    Time will tell which work ends up being useful for future researchers. It’s not always, or even usually, predictable what will be useful. I also think systems will be developed to assist us in filtering out noise.

  2. Hume Avatar
    Hume

    In the future, contributions have two necessary components. One, you need a proof with lean certificate (submitted to Arxiv). Second, you need a YouTube video in which you personally explain the result to the audience (some already suggested requiring videos for journal submissions). This is similar to SWEs having a github that shows what they have done.

    1. Manuel Avatar
      Manuel

      I do not believe that YouTube videos alone will make that big of a difference, but certainly I have no doubt (and agree with you in that essence) that good and useful video recordings can have a positive impact. But there can also be a video-script-mill, managing to script videos to meet some minimal requirements for certain publication venues. I would rather believe an existing mechanism will become even more relevant: in-person or live online seminar talks, lectures, and mini-courses, where the author of a study interacts with expert audiences and mixed audiences. This was already a common method to popularize one’s work during peer-review stage (in the good old days, aka ”before ~July 2026”) and commonly influenced success with publishing in a leading journal: presenting a result in an understandable and motivating way to those who may be involved in peer-review or editorial process undoubtedly can have impact. Moreover, already in the past, work that was presented during various seminars at credible institutions undoubtedly looked more credible and was likely to attract follow up work, thus again increasing chances for publication. What I believe will happen in the nearer future (but I may be wrong) is that this aspect of advertising and presenting ones works within expert communities will increasingly be used to filter out content to be published.

      I am not sure if a lean certificate will become a requirement, maybe it will, maybe just for some journals. One question is, who is really interested in reading and understanding at best thousands of lines of lean? A valid certificate could arise from the exploitation of bugs, which an LLM may be able to utilize and to obscure in long unreadable lean. Certainly, humans regularly make errors as well, but in general they do not try to hide them on purpose, so an honest write up by a human expert may still carry the trust readers actually seek more than a certificate from an oracle. Another issue is, even if one could ensure that LLMs do not find a way to produce a false positive lean certificate, to make this an requirement one would need to bring strong LLM access to all countries and institutions. Maybe journals could run the LLMs that produce the lean with author interaction after conditional acceptance of a manuscript. But the resources needed for this are likely beyond the current scope.

    2. Orr Shalit Avatar

      Lean certificates should not (and probably will not) become a necessary component for submissions of mathematical papers. This will only distance math from humanity while improving very little if anything at all.

      First, a lean certificate for what? Who checks that the formalization formalizes the statement of the paper? Maybe formalization can be formalized? Surely it can. So maybe require a certificate for that too? Surely there must be a Godel theorem for formalization….

      Second, I find it very strange to engineer a change in our epistemological stance of how we decide on mathematical truth. If you told me that a paper that no human mathematician understand requires a lean certificate, I’ll agree. But to follow this logic to the extreme, we now need lean certificates to accept and use results in 20th century papers, at least if we are to invoke these results in our future papers, because for some reason a technological changes has made us stop believing in their truth.

      Finally, I don’t deny that one could make a reasonable argument in favor of the idea of requiring a lean certificate, but the improvement of AI tools renders it obsolete rather than a necessity. Authors can deploy a dozen virtual referees to their papers before submission to check for gaps and mistakes (and if some of the virtual referees decides to use lean – good for them). Journals, on the other hand, have no need for an external certificate (which they might need to check anyway) when they can easily automatically challenge papers for correctness or completeness. One can imagine the appearance of automated mathematical goalkeepers, which are trained to find possible gaps and mistakes in proofs (which human referees can then look into). Arguably, mathematical goal keeping seems to me to be an easier task than automated theorem proving.

      As for videos, I also don’t see it becoming a necessity for journal publications, but it certainly will add something.

      Journals should keep track of submissions and penalize authors for submitting junk or bulk.

  3. Giovanni Mongardi Avatar
    Giovanni Mongardi

    I think the true point raised by this post is that it is not possible to claim authorship of any result in a world where AI proves them: it would be impossible to distinguish between a behaviour all of us consider unacceptable (buying a finished paper from an external party, and then maybe rewriting it and learning its content), and a behaviour some consider acceptable (prompting a known question to an LLM, and then understanding the answer and writing it down)

  4. John Baez Avatar

    One computer science conference proceedings editor recently interviewed ten authors of suspicious papers, to see how well they understood their claimed work. Nine failed the test. While this is not cheap to implement on a large scale, perhaps journals should actually go back to doing some work – having gotten authors to typeset their own manuscripts.

  5. ch Avatar
    ch

    Why would you pay for a paper if no one cares anymore about publication list?

Comments are moderated. Read our comment policy.

Add to the discussion

New posts by email.

Prefer a feed reader? Subscribe by RSS.

Also on Mathstodon.

Latest comments across the site.