Why ICLR Limits Authors to 20 Papers – A DeepMind Researcher’s Satirical Petition

ICLR announced that starting with the 2027 conference, any author may appear on at most 20 paper submissions; exceeding this quota triggers a warning during abstract submission and, if not resolved by the full‑paper deadline, the conference will randomly reject papers until the limit is met.

The rule is explained as a response to a rapid increase in submissions that has strained the review system, requiring a quota to preserve fair and informative reviewing. In 2026, ICLR received 19,525 valid submissions—about a 68% rise over the 11,603 submissions in 2025—resulting in 13,763 papers receiving final decisions and 76,139 review comments from 18,054 reviewers.

DeepMind researcher and University of Toronto professor Dan Roy reacted on X with a petition demanding the removal of the 20‑paper cap. His post sarcastically argues that AI agents can already produce research with minimal guidance and that limiting submissions will “slow down artificial intelligence development.” He flips ICLR’s rationale, pretending to support a flood of incremental papers while ICLR seeks to curb such a flood.

Roy also mocks the claim that “all review work is now handled by large language models,” noting that AI is already present in both writing and reviewing stages. In late 2025, during the ICLR 2026 review peak, Carnegie Mellon AI researcher Graham Neubig used the AI‑text detection tool EditLens (developed by Pangram Labs) to scan publicly released review comments. Out of roughly 75,800 reviews examined, about 21% were flagged as highly likely to be fully AI‑generated.

The detection results sparked industry alarm about declining review quality and potential AI misuse. Roy’s petition, while satirical, highlights a deeper concern: if agents can mass‑produce papers, reviewers can mass‑evaluate them with AI, a simple paper‑count cap may not address the underlying challenges of AI‑augmented research workflows.

The article invites readers to share their views on ICLR’s policy and Roy’s petition, emphasizing the broader debate on how conferences should adapt to the accelerating role of AI in research production and evaluation.