How many papers is it appropriate for one author to submit to a top-tier AI conference?
ICLR provided an answer of 20 articles.
Recently, ICLR 2027 announced new submission guidelines, explicitly stating that any author may appear on at most 20 submitted papers. Papers exceeding this limit will receive a notification during the abstract submission stage. If the excess is not corrected by the full paper deadline, the conference will randomly reject some papers until all authors are within the limit...
ICLR's explanation is straightforward: the number of submissions has grown too rapidly, placing immense pressure on the review system; quotas are needed to maintain fair and informative reviews.
Behind this lies the "crisis" of submission and review experiences from ICLR 2026.
According to the data, ICLR 2026 received 19,525 valid submissions, and final acceptance or rejection decisions were issued for 13,763 papers, supported by 76,139 review comments provided by 18,054 reviewers.
Compared to the approximately 11,603 submissions to ICLR 2025, submissions for 2026 increased by about 68%. This means that within a single year, the conference must process nearly 8,000 additional papers and tens of thousands of additional reviews...
Upon announcement, the news sparked widespread discussion among netizens. Dan Roy, a visiting researcher at Google DeepMind and professor at the University of Toronto, also noticed this change and launched a petition calling for the removal of the limit of 20 paper submissions per author at the ICLR conference.
He posted on X:
ICLR limits each author to 20 submissions, which is absurd! It will also slow down the development of artificial intelligence.
Why exactly 20? It’s completely an arbitrary number. Today, AI agents can already perform interesting but incremental AI research with minimal guidance. We should encourage everyone to complete and submit as many papers as possible, especially now that all peer review is handled by large language models… Admit it!
Click the link to sign my petition!
Clearly, Dan Roy is dissatisfied with the 20-article limit, but is also subtly mocking the current impact of AI on academia. The entire post carries an ironic tone.
For example, ICLR reminds researchers that current AI agents are still unable to independently complete a paper meeting top conference standards. After code assistants and research infrastructure improve efficiency, researchers should tackle more ambitious problems and submit more complete, mature results, reducing the influx of numerous incremental papers into conferences.
Dan Roy, on the other hand, completely reversed this wording.
ICLR wants to see a more complete "slow science"; Dan Roy pretends to advocate for "letting agents produce as many incremental results as possible"; ICLR tries to limit the flood of papers, and he pretends to worry that "this will slow down the development of artificial intelligence."
In particular, the remark “Now all review work has been handed over to large language models… admit it” is also satirical, highlighting that AI is not only present in the paper writing process but also in the peer review process.
Dan Roy was not exaggerating this point.
By the end of 2025, during the peak review period for ICLR 2026, renowned AI researcher Graham Neubig at Carnegie Mellon University noticed anomalies after receiving several review comments, suspecting AI-generated "filler." He then used EditLens, an AI text detection tool developed by Pangram Labs, to scan and analyze the publicly available conference reviews.
The results showed that, after analyzing 75,800 reviews, approximately 21% were highly suspected to be "entirely generated by AI."
The news sent shockwaves through the industry and sparked concerns in academia about declining review quality and the "misuse of AI."
Perhaps, in this "petition," what Dan Roy truly wants to raise is that as the scenario of agents generating papers in bulk, researchers submitting papers in bulk, and reviewers using large models to conduct reviews in bulk becomes increasingly common, a blunt, one-size-fits-all approach—merely limiting the number of paper submissions—may not be the best solution...
What about you—what are your thoughts on ICLR’s actions and Dan Roy’s petition? Feel free to leave a comment and share your views!
Reference link:
https://iclr.cc/Conferences/2027/AuthorGuidelines
https://x.com/roydanroy/status/2083202134414102594
This article is from the WeChat public account "Machine Heart" (ID: almosthuman2014), authored by someone interested in AI.