[Hotstorage-sc] Input regarding AI generated submissions to HS '26

[Ashvin Goel]

The 10 papers were submitted by the same set of authors.

Ashvin

On Tue, Jul 14, 2026 at 8:28 PM Erez Zadok <ezk at fsl.cs.sunysb.edu> wrote:

> That is a hotly debated topic nowadays among ACM EiCs and multiple
> conferences.  We just had heated debates about this on the ACM ATC SC too.
> Different people are trying different things to varying degrees of success,
> from requiring disclosure by authors, to automated tools, etc.  No perfect
> solutions.  ACM seems to have given up on getting an AI-detection tools but
> are looking into cite-checking tools.
>
> ACM recently loosened their AI policy: they no longer prohibit it outright
> but ask authors to disclose AI use if wanted, then ask the reviewers to
> decide what, if, and how much AI use is ok or not ok.  IOW, ACM now shifted
> the burden onto reviewers.  Some are ok with this change, others less so.
>
> One problem with any AI detectors is that actual text can easily be
> flagged as AI-generated even if the authors use AI to spell check, polish
> the text, or ask AI to shrink the text size: these are considered valid
> uses of AI that shouldn’t result in rejection.
>
> That said, the easiest way to catch bad AI use is via citation checks.  At
> TOS, we now validate each citation of a submission before sending it off to
> be reviewed.  Even with search tools, that’s more work but we have no
> choice: recently we caught several papers with bad/hallucinated cites deep
> into the reviewing process (thus wasting reviewers’ precious time).  Any
> paper with obviously bad/hallucinated cites gets desk-rejected, but we also
> email the authors to see if they can explain themselves first.
>
> So my suggestions are:
>
> 1. Papers with obviously large portions of AI text should be rejected with
> minimal reviews.  If authors didn’t spend any significant effort to develop
> the paper, then our reviewers shouldn’t waste their time either.  “You get
> what you paid for” principle.
>
> 2. Papers with too many hallucinated cites should be rejected right away
> without review.  1-2 cites max may be considered a silly mistake that
> students sometimes make (I’d ask the authors), but any more should be
> rejected.
>
> 3. I don’t think we should limit the number of submissions per author.
> While rare, it’s possible to see 2-3 papers submitted by a subset of the
> same authors.  Now, when you say “a set of authors had submitted 10
> papers,” *how many* are in the set?!  Is it the same exact authors who
> submitted 10 separate papers?
>
> Thanks,
> Erez.
>
> > On Jul 14, 2026, at 2:18 PM, Ashvin Goel via Hotstorage-sc <
> hotstorage-sc at fsl.cs.sunysb.edu> wrote:
> >
> > Hi all,
> >
> > The HS '26 PC chairs (Young-ri and Bryan) are asking for input from us
> regarding a set of paper submissions that have likely been AI generated.
> Two reviewers expressed concerns to the chairs. They found several papers
> with similar concepts applied to different areas, following a similar
> structure, language (including hard to understand or strange words) and
> flaws. The chairs found that a set of authors had submitted 10 papers (13%
> of the 79 submissions). The reviewers are currently leaning negative on
> several of these papers, pointing out that the papers are difficult to
> understand, describe concepts vaguely, and use unfamiliar language. The
> positive-leaning papers will be discussed at the PC meeting.
> >
> > The chairs are looking for input about the best course of action to
> handle this case of suspected AI-generated papers. On the one hand, ACM
> allows the use of undisclosed AI. Thus, rejections should be grounded on
> the paper's technical contribution and novelty, rather than AI use, which
> is hard to detect accurately using current AI detection tools. On the other
> hand, such AI-generated papers will overwhelm reviewers in the future.
> >
> > Let me know if you have any specific suggestions regarding handling the
> current case with the 10 submissions. More broadly, we should discuss if HS
> should put in place any policies for AI generated papers. Here are some
> examples I have seen:
> >
> > 1. Using multiple detectors and letting reviewers know so they can decide
> > 2. Desk rejecting papers with hallucinated references
> > 3. Limiting number of submissions by authors
> >
> > Thanks
> > Ashvin
> > --
> > Hotstorage-sc mailing list
> > Hotstorage-sc at fsl.cs.sunysb.edu
> > https://linkprotect.cudasvc.com/url?a=https%3a%2f%2flists.fsl.cs.sunysb.edu%2fmailman%2flistinfo%2fhotstorage-sc&c=E,1,LtXzFWsPqWpocyejbJE3YlUk3-xm3BPSaKVF0WDDL5Q7hbI6xBtx8BSGyGY--6q2SL_7Gp6L7tQedprs770K4IqXlMwuKJ42j5cOCk7uoc1dlBpqXA,,&typo=1
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fsl.cs.sunysb.edu/pipermail/hotstorage-sc/attachments/20260714/ac957462/attachment.htm>