20260710 - NeurIPS conference AI cheats outraged they got caught

5 min

•Jul 10, 202618 days ago

Summary

NeurIPS and ICML conferences caught reviewers using AI chatbots to write peer reviews despite explicit policies forbidding it. The conferences embedded hidden prompts in PDFs to detect violations, and caught hundreds of cheating reviewers whose own papers were rejected as punishment. While some reviewers complained on LinkedIn, the scientific community overwhelmingly supported the enforcement strategy.

Insights

Academic peer review integrity is under threat from reviewers outsourcing work to LLMs, undermining the entire scientific validation process
Detection methods using embedded prompts in PDFs are effective at catching policy violations and can serve as a model for other conferences
There is a significant generational or cultural divide between AI industry professionals and academic researchers on ethical use of AI tools
Enforcement with consequences (paper rejection) is necessary and effective—the scientific community strongly supports accountability over trust-based policies
Explicit opt-in policies (like ICML's dual-track approach) may be more effective than blanket bans at managing AI tool use in peer review

Trends

AI ethics enforcement becoming standard practice at major academic conferencesGrowing tension between AI industry culture and academic scientific integrity standardsDetection and accountability mechanisms replacing honor-system policies in peer reviewOverwhelming researcher support for strict AI tool restrictions in peer review processesDual-policy approaches (permissive and restrictive tracks) emerging as compromise solutionsPeer review quality degradation from copy-pasted AI-generated reviews becoming a recognized problemConferences implementing technical solutions (embedded prompts) to enforce policy compliance

Topics

Peer review integrity and AI chatbot misuse NeurIPS conference policies and enforcement ICML dual-track policy approach LLM detection methods in academic publishing Academic ethics and AI tool governance Reviewer accountability and consequences Machine learning conference standards Scientific community sentiment on AI in peer review Policy compliance monitoring techniques AI industry culture vs. academic culture clash

Companies

NeurIPS

Major annual machine learning conference that implemented hidden chatbot detection prompts in review PDFs to catch po...

ICML

International Conference on Machine Learning that pioneered dual-track peer review policy allowing optional LLM use a...

People

David Girard

Host of the Pivot to AI podcast discussing the NeurIPS and ICML peer review cheating scandal

Shah

ICML researcher who shared the chatbot detection methodology with NeurIPS and noted overwhelming support from the sci...

Quotes

"Our policy is strict. Reviewers may not use any LLMs or AI agents in the review process"

NeurIPS policy statement•~2:30

"You do not build a healthy reviewing culture by treating your reviewers as suspects"

Caught reviewer on LinkedIn•~3:45

"795 reviews, about 1% of all reviews, written by 506 unique reviewers who were assigned policy A, no LLMs, were detected to have used LLMs in their review"

ICML enforcement report•~5:00

"Researchers expressed overwhelming support for the strategy... People were really tired of reviewers copy-pasting AI-generated reviews without putting in any effort"

Shah•~6:30

Full Transcript

Hi, I'm David Girard and this is Pivot to AI, coming to you Monday to Friday. Today, nobody likes to cheat. Neuro IPS is an annual conference on machine learning. Here in the AI bubble, Neuro IPS is huge, and these days it's flooded with chatbot cash. The AI vendors at NeurIPS want to talk about imaginary problems like the super AI destroying humanity. The researchers want to get on with doing science. So, you want to present your work at NeurIPS. You submit your conference paper and it goes to a reviewer, usually someone else who submitted a paper. They check your papers up to standard. That's how peer review works. Unfortunately, a lot of reviewers these days pass their reviewing to the chatbot, which is missing the entire point of peer review. New IPS has a clear rule. It flatly forbids reviewers from using the chatbot. Quote, Our policy is strict. Reviewers may not use any LLMs or AI agents in the review process Unquote But a few reviewers didn listen and now they yelling in such august scientific venues as LinkedIn that they got caught following their work off to a chatbot. How did they get caught? NeurIPS put chatbot prompts hidden in the PDFs they send out for review. These tell the chatbot to use a particular phrase in the review. The bros were deeply aggrieved. Quote, You do not build a healthy reviewing culture by treating your reviewers as suspects. Unquote. Said the guy who'd just been busted feeding a paper to the bot. Neuro-IPS was not the first machine learning conference to try this trick. The International Conference on Machine Learning did the same thing earlier this year. ICML actually had two policies for chatbot reviewing. Policy A was no chatbots, no chatbots allowed. Policy B was, quote, LLMs allowed to help understand the paper and related works and polish reviews Unquote Fine That easy But AI bros refuse to understand consent. They seem offended when they have to consent. So a pile of reviewers signed up for the no chatbots track, and then they used the chatbots anyway, and they got busted. Quote. 795 reviews, about 1% of all reviews, written by 506 unique reviewers who were assigned policy A, no LLMs, were detected to have used LLMs in their review. Again, recall that these are reviewers who explicitly agreed to not use LLMs in their reviews. If the designated reciprocal reviewer for a submission produced such a review, their submission was rejected. Unquote. So the bros who cheated got their own paper kicked out. How did this go down? Was there a flood of deep concern? It turns out the AI bros whining on LinkedIn are not the majority. Their scientists were delighted with the ICML approach Quote Researchers expressed overwhelming support for the strategy says Shah who adds that he shared the methodology with the new IPS team. I've been working on conference peer review for several years, and I've hardly seen such strong support for anything, he says. People were really tired of reviewers copy-pasting AI-generated reviews without putting in any effort. Unquote. Academic reviews are work, but nobody likes a cheat. Thanks for tuning in to Pivot to AI. Please do forward this episode to all the scientists you know. I bet they've got some things to say. Hit like and subscribe on YouTube. Leave a nice review in your podcast app. And don't forget, Pivot to AI runs on the support of our viewers and listeners. Please do put your $5 into the Patreon linked in the show notes. We need this to keep doing this. Thank you all. I'll see you on Monday and bye for now.