Major conference catches illicit AI use — and rejects hundreds of papers

## Summary
Email Bluesky Facebook LinkedIn Reddit Whatsapp X Organizers of the 2026 International Conference on Machine Learning (ICML) used a watermarking system to catch the use of AI in peer review of conference papers. The International Conference on Machine Learning (ICML), to be held in Seoul in July, has a reciprocal review policy, meaning that, bar certain exceptions, every paper must have an author who reviews other conference papers. PLoS ONE 20 , e0331871 (2025) Article PubMed Google Scholar Download references Reprints and permissions Related Articles More than half of researchers now use AI for peer review — often against guidance Bring us your LLMs: why peer review is good for AI models Scientists hide messages in papers to game AI peer review The peer-review crisis: how to fix an overloaded system AI, peer review and the human activity of science This AI can improve your peer review — and make it more polite Subjects Computer science Machine learning Peer review Latest on: Computer science Machine learning Peer review These medical X-rays are all deepfakes — and they fool even radiologists News 24 MAR 26 UK bets big on homegrown fusion and quantum — can it lead the world? News 19 MAR 26 Major Turing computing award goes to quantum science for first time News 18 MAR 26 Why AI hasn’t caused a job apocalypse — so far Comment 24 MAR 26 These medical X-rays are all deepfakes — and they fool even radiologists News 24 MAR 26 Rethinking AI’s role in survey research: from threat to collaboration Correspondence 17 MAR 26 Five ways to spot when a paper is a fraud Technology Feature 25 FEB 26 This AI can improve your peer review — and make it more polite News 23 FEB 26 How AI slop is causing a crisis in computer science News 13 FEB 26 Jobs Postdoctoral Fellow A postdoctoral fellow position is available at NIH for translational research on neural mechanisms of frustration/irritability in mice and humans Bethesda, Maryland (US) National Institute of Mental Health, National Institutes of Health Junior Professorship (W1) for „Public Child Mental Health“ (f/m/d The Medical Faculty Mannheim of Heidelberg University invites applications for the Junior Professorship (W1) for “Public Child Mental Health” (f/...

## Article Content
Email
Bluesky
Facebook
LinkedIn
Reddit
Whatsapp
X
Organizers of the 2026 International Conference on Machine Learning (ICML) used a watermarking system to catch the use of AI in peer review of conference papers.
Credit: Stephen Barnes/Alamy
A major artificial-intelligence conference has rejected 497 papers — roughly 2% of submissions — whose authors violated AI-use policies in their peer reviews of other articles submitted to the meeting.
The International Conference on Machine Learning (ICML), to be held in Seoul in July, has a reciprocal review policy, meaning that, bar certain exceptions, every paper must have an author who reviews other conference papers. Authors whose reviews violated the conference’s large language model (LLM)-use policy had their papers rejected.
Conference organizers detected the illicit AI use by hiding watermarks in research papers distributed for review. If a researcher used an LLM to generate their peer review, instructions hidden in the watermark prompted the LLM to include telltale phrases in the review text. The presence of these phrases revealed that an AI model had been used to generate the review.
“We hope that by taking strong action against violations of agreed-upon policy we will remind the community that as our field changes rapidly the thing we must protect most actively is our trust in each other,” said the organizers in
a blog post on 18 March
.
“What the ICML case shows is a research community in need of clear guidance on responsible AI use, including use in peer review,” says Marie Soulière, head of editorial ethics and quality assurance at the publishing company Frontiers in Lausanne, Switzerland.
More than half of researchers now use AI for peer review — often against guidance
Many researchers applauded the news of the ICML’s actions
on the social-media platform X
. Some suggested that other conferences could learn from the ICML or that the meeting could go further, for example by banning the rejected authors from resubmitting.
However, Zhengzhong Tu, a computer scientist at Texas A&M University in College Station, said the policy would be ineffective. “It will only demotivate all the reviewers,” he said, adding that they will avoid routes banning AI use and will use LLMs “to generate meaningless reviews”.
AI reviewer
Using AI in peer review is now common:
more than half of researchers in a 2025 survey from Frontiers have
done it, despite journals and conference policies often banning its use. But AI researchers are starkly divided on whether LLMs should have a role in peer review.
This sparked ICML organizers, for the first time, to operate two peer-review streams — one allowing limited LLM use, and the other strictly prohibiting it. Authors and reviewers were allowed to choose their preferred stream.
Enjoying our latest content?
Log in or create an account to continue
Access the most recent journalism from Nature's award-winning team
Explore the latest features & opinion covering groundbreaking research
Access through your institution
or
Sign in or create an account
Continue with Google
Continue with ORCiD
doi: https://doi.org/10.1038/d41586-026-00893-2
References
Rao, V. S., Kumar, A., Lakkaraju, H. & Shah, N. B.
PLoS ONE
20
, e0331871 (2025)
Article
PubMed
Google Scholar
Download references
Reprints and permissions
Related Articles
More than half of researchers now use AI for peer review — often against guidance
Bring us your LLMs: why peer review is good for AI models
Scientists hide messages in papers to game AI peer review
The peer-review crisis: how to fix an overloaded system
AI, peer review and the human activity of science
This AI can improve your peer review — and make it more polite
Subjects
Computer science
Machine learning
Peer review
Latest on:
Computer science
Machine learning
Peer review
These medical X-rays are all deepfakes — and they fool even radiologists
News
24 MAR 26
UK bets big on homegrown fusion and quantum — can it lead the world?
News
19 MAR 26
Major Turing computing award goes to quantum science for first time
News
18 MAR 26
Why AI hasn’t caused a job apocalypse — so far
Comment
24 MAR 26
These medical X-rays are all deepfakes — and they fool even radiologists
News
24 MAR 26
Rethinking AI’s role in survey research: from threat to collaboration
Correspondence
17 MAR 26
Five ways to spot when a paper is a fraud
Technology Feature
25 FEB 26
This AI can improve your peer review — and make it more polite
News
23 FEB 26
How AI slop is causing a crisis in computer science
News
13 FEB 26
Jobs
Postdoctoral Fellow
A postdoctoral fellow position is available at NIH for translational research on neural mechanisms of frustration/irritability in mice and humans
Bethesda, Maryland (US)
National Institute of Mental Health, National Institutes of Health
Junior Professorship (W1) for „Public Child Mental Health“ (f/m/d
The Medical Faculty Mannheim of Heidelberg University invites applications for the   Junior Professorship (W1) for “Public Child Men

---

## Expert Analysis

### Merits
- The presence of these phrases revealed that an AI model had been used to generate the review. “We hope that by taking strong action against violations of agreed-upon policy we will remind the community that as our field changes rapidly the thing we must protect most actively is our trust in each other,” said the organizers in a blog post on 18 March . “What the ICML case shows is a research community in need of clear guidance on responsible AI use, including use in peer review,” says Marie Soulière, head of editorial ethics and quality assurance at the publishing company Frontiers in Lausanne, Switzerland.

### Areas for Consideration
- News 19 MAR 26 Major Turing computing award goes to quantum science for first time News 18 MAR 26 Why AI hasn’t caused a job apocalypse — so far Comment 24 MAR 26 These medical X-rays are all deepfakes — and they fool even radiologists News 24 MAR 26 Rethinking AI’s role in survey research: from threat to collaboration Correspondence 17 MAR 26 Five ways to spot when a paper is a fraud Technology Feature 25 FEB 26 This AI can improve your peer review — and make it more polite News 23 FEB 26 How AI slop is causing a crisis in computer science News 13 FEB 26 Jobs Postdoctoral Fellow A postdoctoral fellow position is available at NIH for translational research on neural mechanisms of frustration/irritability in mice and humans Bethesda, Maryland (US) National Institute of Mental Health, National Institutes of Health Junior Professorship (W1) for „Public Child Mental Health“ (f/m/d The Medical Faculty Mannheim of Heidelberg University invites applications for the Junior Professorship (W1) for “Public Child Mental Health” (f/...

### Implications
- The International Conference on Machine Learning (ICML), to be held in Seoul in July, has a reciprocal review policy, meaning that, bar certain exceptions, every paper must have an author who reviews other conference papers.
- Authors whose reviews violated the conference’s large language model (LLM)-use policy had their papers rejected.
- The presence of these phrases revealed that an AI model had been used to generate the review. “We hope that by taking strong action against violations of agreed-upon policy we will remind the community that as our field changes rapidly the thing we must protect most actively is our trust in each other,” said the organizers in a blog post on 18 March . “What the ICML case shows is a research community in need of clear guidance on responsible AI use, including use in peer review,” says Marie Soulière, head of editorial ethics and quality assurance at the publishing company Frontiers in Lausanne, Switzerland.
- Some suggested that other conferences could learn from the ICML or that the meeting could go further, for example by banning the rejected authors from resubmitting.

### Expert Commentary
This article covers review, peer, conference topics. Notable strengths include discussion of review. Areas of concern are also raised. Readability: Flesch-Kincaid grade 0.0. Word count: 917.