Post #2901584

2026-05-16 11:05 UTC

You might have read that arxiv is banning people for a year if they post LLM-generated papers and cheered it on. But most of the discussion about this doesn&#39;t correctly explain the policy and it is not a good thing. First up, the policy is that &quot;incontrovertible evidence&quot; of using LLMs and not checking the output is what&#39;s at stake. An example given is a hallucinated reference. Second, the ban will apply to all coauthors of the paper, not just the person submitting. Third, it&#39;s not just a 1 year ban, it&#39;s followed by a permanent ban on submitting papers that have not been peer reviewed in a &quot;reputable&quot; journal or conference. Given that arxiv is a preprint server and not a repository for published papers, that makes the ban effectively permanent. (EDIT: They&#39;ve subsequently clarified that this isn&#39;t a permanent ban but will be lifted after 1 or 3 (the clarification isn&#39;t clear) peer reviewed papers. This is still problematic, but much better. The rest of this post left as I originally wrote it.) So imagine: you are a masters student working on a project with a few other people, and your role is relatively minor. The project leads to a paper and you get your name on it, hurray. The lead author handles the submission and doesn&#39;t ask for your permission to send the final version because you&#39;re only a masters student. Your supervisor explains that this is how things are done and nothing to worry about. What you didn&#39;t know is that someone else on this paper at the last minute made some edits to the grammar of the paper using an LLM because none of you are native English speakers, and the LLM inserted a hallucinated reference. Arxiv picks up on this and you are now permanently banned from using arxiv as a preprint server. Further, every time you try to collaborate with someone else to write a paper and they want to put it in arxiv you have to explain that you can&#39;t, and that this means that by collaborating with you, they also can&#39;t put it on arxiv. Since arxiv is one of the main channels for distributing papers in your field, soon enough people stop asking you to work with them and your career is effectively over. Because someone else didn&#39;t notice that an overenthusiastic grammar checker inserted a fake reference and you weren&#39;t in a position of enough power at the time to insist on checking the final version. Ok that&#39;s a long story, but I don&#39;t think this is a fanciful situation. Stuff like this happens all the time. It&#39;s easy to say - and I&#39;ve seen a lot of people saying times like this - that everyone should take responsibility for reading the paper, or that it&#39;s the responsibility of supervisors to make sure this doesn&#39;t happen. But in the world we actually inhabit, power imbalances exist: the masters student can&#39;t make the supervisor wait until they read the paper because they&#39;re worried about their project grade. Bad supervisors are out there, and it&#39;s not fair to punish their students. This policy will lead to terrible consequences for a lot of innocent people who should not reasonably be held responsible because they weren&#39;t in a position of power. I suspect it won&#39;t lead to very bad consequences for big name researchers who will just get on the phone to someone at arxiv or one of arxiv&#39;s funders and get the decision reversed in their case. I understand the anger towards LLMs and tech companies, and I share it. I understand the anger towards the people cynically generating whole papers using them, polluting the scientific literature and making all our lives more difficult, and I share it too. But that doesn&#39;t mean we should jump to implement extreme and poorly thought out policies that will hurt a lot of people who haven&#39;t done anything wrong. Finally, as an advocate of open science and publishing reform, this is really disappointing from arxiv. By saying that peer reviewed papers in &quot;reputable&quot; journals are ok, they&#39;ve defined themselves (arxiv) as second class citizens in the world of publishing. This shows such limited ambition, and actively hurts the cause of making the world better by getting rid of the parasitic and harmful publishing industry. #academicchatter #arxiv

Replies (15)

@steveroyle@biologists.social 2026-05-16 11:53

@neuralreckoning@neuromatch.social I can't figure out whether this is a real policy or not. So far there's an X thread and a couple of Bluesky comments by someone running a subgroup of arXiv. The submission guidelines don't mention it. I completely agree with your take though (if the policy is implemented).

Open ##3286771
@adredish@neuromatch.social 2026-05-16 12:06

@neuralreckoning@neuromatch.social I do think they have an appeal process, but it's not clear how that actually works. #arXiv has (for a while) had a policy that they will only "preprint"* certain kinds of papers after they appear in a "peer reviewed journal". We ran into this buzzsaw on a theory/perspective paper last year. (No LLMs involved.) Our paper ended up preprinted on psyArxiv, which didn't have this weird rule. * How is it a preprint if it's already been published? I suppose there's a problem of how to navigate the flood of slop that's now in the system, but it definitely seems that arXiv is struggling with finding a good policy on this.

Open ##3286773
@smiergahttu@fosstodon.org 2026-05-16 12:59

@neuralreckoning@neuromatch.social The rule doesn't seem to be exaggerated, considering the problem with generated papers. Especially since it is not LLMs that are banned, but only unchecked, not revised, hallucinated parts. So it is like "just read a paper carefully before uploading. Please?"

Open ##3286780
@ahltorp@mastodon.nu 2026-05-16 16:33

@neuralreckoning@neuromatch.social How are authors identified on arXiv? Can someone just add an author and have them banned? And if the author has to be involved in submitting the paper through some verification process, why is that verification process not applied for the final version? For conference submissions, there are deadlines making this often be unworkable, but why should a preprint service not enforce final approval by all authors?

Open ##3286781
@hyc@mastodon.social 2026-05-16 18:05

@neuralreckoning@neuromatch.social aren't there still grammar checkers that don't use LLMs? If you know going in that the site has zero tolerance for LLM "hallucinations", it's on you to proof read. I see nothing to disagree with here.

Open ##3286785
@grayrattus@mastodon.social 2026-05-16 19:38

@neuralreckoning@neuromatch.social I think this is the game changer policy which will increase quality of papers. "What you didn't know is that someone else on this paper at the last minute made some edits to the grammar of the paper using an LLM" - but you use LaTeX for your paper and store in git right? If now how do you make sure your coleagues don't write trash? You review whole paper every day? I read your arguments and I think example you given is some edge case which probably will never happen. #science

Open ##3286786
@drgroftehauge@sigmoid.social 2026-05-16 23:22

@neuralreckoning@neuromatch.social And fines for a company breaking the law should never be larger than what can be easily absorbed on the balance sheet - otherwise an innocent person could lose their job. Your assertion that arXiv leadership is deeply corrupt is really interesting though.

Open ##3286788
@jonny@neuromatch.social 2026-05-17 00:08

@neuralreckoning@neuromatch.social I think that this is pretty easily resolvable with an appeals process that is sensitive to power imbalances like this. Not like I think arxiv's moderation is always fair or aligned with my interests, but pretty much all AI policies require editorial discretion, and I'd rather a strong policy with discretion than a weak policy that leaves the door open for poisoning the entire concept of preprints.

Open ##3286791
@bugaevc@floss.social 2026-05-17 04:29

@neuralreckoning@neuromatch.social how does grammar checking result in a hallucinated paper reference?

Open ##3286795
@julian_kuelshammer@mathstodon.xyz 2026-05-17 06:43

@neuralreckoning@neuromatch.social They have since clarified that the ban is not permanent, but only the first few papers will have to be published first. Also, they announced that some changes to the submission system are coming. They didn't give all the details, but it sounded like more approval from coauthors.

Open ##3286797
@f4grx@chaos.social 2026-05-17 07:48

@neuralreckoning@neuromatch.social this is the correct way. Render the use of extruded vomit socially unacceptable. It's unfair? Good. These students will remember to not work with these sloppers anymore. Slop is 100% unacceptable in the academic research domain. It's a disgrace to scientific knowledge itself.

Open ##3286803
@renebekkers@mastodon.social 2026-05-17 08:39

@neuralreckoning@neuromatch.social thank you for going through the details of this policy

Open ##3286804
@MCDuncanLab@mstdn.social 2026-05-17 12:01

@neuralreckoning@neuromatch.social Thank you for this detailed description. I it sounds like arxiv has gone a bit too far. One hopes community pushback will be successful in changing their minds.

Open ##3286805
@mastodonmigration@mastodon.online 2026-05-17 14:55

@neuralreckoning@neuromatch.social This argument is fundamentally flawed. Imagine a frat party where a young woman is raped by members of the football team. Should those who 'just watched' be let off the hook? What about those who refused to watch, but didn't report the crime? The purpose of the policy is to make every person who affixes their name to a paper personnally and independently attest it is free of slop. If one is not sure, it is incumbent upon the individual to have their name removed. Simple.

Open ##3286806
@olibrendel@scicomm.xyz 2026-05-18 15:38

@neuralreckoning@neuromatch.social do you have a link for this information please ?

Open ##3286807