News from Science reports that the International Conference on Learning Representations received more than 40,000 paper submissions for its 2027 meeting. About a decade ago, the leading AI conference received roughly 500.
The increase is landing on a peer-review system that already handles an enormous workload. ICLR’s retrospective for 2026 counted 19,525 valid, format-compliant submissions. After withdrawals and procedural rejections, 13,763 papers received decisions based on 76,139 reviews from 18,054 reviewers.
The numbers need context
An independent analysis of OpenReview records found more than 62,000 registered abstract IDs for 2027. That is not a verified count of valid full-paper submissions: ICLR removes placeholders and duplicates, and some authors withdraw before review. The more conservative figure in the Science report refers to papers submitted after the deadline.
News from Science says researchers point to several forces behind the broader surge, including intense interest in AI, pressure to publish and tools that make paper writing faster. Those explanations describe a mix of incentives and technology, not a single proven cause.
ICLR is rationing reviewer time
For 2027, ICLR’s author rules cap any author at 20 papers. A team with no author who qualifies as a reciprocal reviewer may submit only one paper. Submissions above either limit can be randomly desk-rejected until every author is under the quota.
The conference’s program chairs said the limits are meant to protect review quality and reduce low-quality mass submissions. They also acknowledged trade-offs: junior researchers in large labs could be constrained, while the rule could create incentives to add qualified reviewers as nominal co-authors. ICLR said it plans to monitor for that behavior.
