ICLR 2027 Registrations Hit 60K, Exceeding All Prior Years Combined

ICLR 2027 abstract registration (取号) on OpenReview officially broke 60,000, a figure that already exceeds the cumulative submissions of all previous ICLR conferences from 2013 through 2026, which totaled approximately 55,972 papers. The registration count is not the final submission number: authors must submit full papers a week later, and historical data show significant attrition through withdrawals and desk rejects.

From Registrations to Review Load

ICLR 2026 provides a baseline: 19,525 valid submissions, 779 desk rejects, 5,042 withdrawals, leaving 13,763 papers that completed the full review process. That volume required 18,054 reviewers who produced 76,139 reviews. Meta senior research scientist Mariya Vasileva modeled several scenarios for 2027 based on 2026 attrition rates:

If 50% of 60,000 registrations become full submissions, ~21,146 papers would reach review — a 54% increase over 2026.

At 70% submission rate, ~29,604 papers would be reviewed.

If all 60,000 convert, ~42,292 papers would enter review — over three times the 2026 load.

Even the conservative 50% scenario implies a substantially larger reviewing burden than last year.

Accelerated Paper Production: The "Research Lottery"

The article attributes the submission surge partly to AI tools compressing the research pipeline. Large language models and autonomous research agents now assist with literature review, topic selection, experimental design, code implementation, and manuscript writing, shrinking cycles that once took weeks or months. Researchers can enter new areas faster, combine and repackage existing ideas, and search for novel angles — a dynamic dubbed the "research lottery."

Pangram analyzed 19,490 ICLR 2026 submissions and found approximately 9% of papers contained more than 50% AI-generated content.

AI-Generated Reviews and Conference Responses

The review side faces parallel automation. Pangram detected that roughly 21% of ICLR 2026's 15,899 public reviews were fully AI-generated. In response, the ICLR committee issued a statement on LLM-generated reviews and low-quality reviews, scanning all reviews with two detectors. NeurIPS 2026 tightened rules further: its Position Paper Track prohibited reviewers from using AI to write reviews outright and screened submissions for AI-generated content, resulting in 178 desk rejects and 123 papers asked to demonstrate substantial human involvement.

With ICLR 2027 registrations at 60,000 and reviewer pools unlikely to scale proportionally, the article argues AI-assisted reviewing will likely increase.

Projection and Policy Measures

Extrapolating current growth, one estimate suggests submissions could approach 900,000 by 2031. ICLR 2027 introduced rate-limiting rules: each author may participate in at most 20 submissions; if none of a paper's authors hold reciprocal reviewer status, each author is limited to one such submission. The ICLR call for papers explicitly acknowledged that submission volume is outpacing reviewer supply and urged researchers to use efficiency gains for "more complete, more solid slow science" rather than simply submitting more papers.

References

[1] https://x.com/i/trending/2101364402192822776 [2] https://mariya.fyi/posts/research-lottery [3]

https://blog.iclr.cc/2026/03/31/a-retrospective-on-the-iclr-2026-review-process/

[4]

https://blog.iclr.cc/2026/09/02/submission-policies-for-iclr-2027/

[5] https://www.iclr.cc/Conferences/2027/CallForPapers [6] https://iclr.cc/Conferences/2027/AuthorGuidelines [7]

https://www.pangram.com/blog/pangram-predicts-21-of-iclr-reviews-are-ai-generated

[8] https://blog.iclr.cc/2025/11/19/ [9]

https://blog.neurips.cc/2026/06/02/ai-generated-papers-in-the-neurips-2026-position-paper-track/