ICLR Caps New Authors at One Paper as AI Strains Peer Review

On September 2, 2026, ICLR 2027 introduced two submission quotas: no author may appear on more than 20 papers, and each author may appear on only one paper whose team has no eligible reciprocal reviewer. A September 3 independent report summarized the same limits and the chairs’ argument that AI-assisted production is increasing pressure on scarce expert review capacity.

The “new authors” label is shorthand, not a test of age, seniority or previous attendance at ICLR. The one-paper rule applies when none of a submission’s authors has a qualifying publication by the September 18 abstract deadline; papers still exceeding either quota at the September 25 full-paper deadline can be randomly desk-rejected.

Who is subject to the one-paper rule

The restriction is determined at team level. Each author may appear on at most one submission for which no co-author qualifies as a reciprocal reviewer. Every person on such a paper uses their one permitted no-reviewer slot.

Eligibility requires at least one accepted publication at a venue recognized by ICLR. The list covers specified machine-learning, natural-language-processing, computer-vision, robotics and related conferences and journals, including ICLR, NeurIPS, ICML, UAI, AISTATS, JMLR, TMLR, ACL, CVPR, ICCV, ECCV, AAAI, IJCAI, ICRA, IROS, RSS, CoRL, KDD and COLT. Primary conference papers count, while workshop papers, position papers, tiny papers, blog posts, system demonstrations and industry papers do not.

A sole author without a qualifying publication may therefore submit one no-reviewer paper. A collaboration composed entirely of ineligible authors may also submit a paper, but none of its members may appear on another submission that lacks an eligible reviewer. If one qualifying author joins the team, the paper is outside this restricted category, although every co-author remains subject to the 20-paper ceiling.

How the 20-paper ceiling reaches large collaborations

The general quota counts every appearance on an author list, regardless of author order, project leadership or institutional affiliation. A researcher named on work from several students or laboratories must count all of those submissions toward the same maximum.

The two quotas operate simultaneously. An author without a qualifying publication could appear on several papers when each includes another author who is eligible to review, potentially reaching the general ceiling. That person may still appear on only one paper for which the entire team lacks eligibility.

The chairs describe 20 as a high threshold aimed at extreme volume, not a normal submission target. Their September 2 policy explanation says fewer than 0.2% of authors exceeded 20 submissions in the previous cycle and notes that junior researchers in large groups may not control which papers list a principal investigator.

The compliance path before submissions close

The official ICLR 2027 author guidelines set the abstract deadline for September 18, 2026, and the full-paper deadline for September 25, both at 11:59 p.m. Anywhere on Earth. They also define the quotas, eligibility rules and desk-rejection process.

1. By September 18: submit a genuine abstract and finalize the people on the author list. Qualifying publications are assessed as of this deadline.
2. After September 18: authors cannot be added or removed, although their order may be changed until the full-paper deadline. Adding an established researcher later cannot cure a no-reviewer classification.
3. By September 25: upload the full paper, make sure every author has an OpenReview profile and bring each author’s complete submission portfolio within both quotas.

Exceeding a quota during the abstract period produces a flag rather than an immediate rejection. If the violation remains when full-paper submissions close, ICLR will randomly desk-reject enough excess papers to bring every affected author under the applicable limit. Because one over-limit collaborator may appear on several papers, the risk extends across that person’s submission portfolio.

The quotas do not replace reciprocal-review duties. Each submission normally needs at least one qualified author registered to review at least three papers. A team with no qualified author is exempt from that requirement but receives the one-paper restriction. Authors appearing on three or more submissions must ordinarily review at least six papers; incomplete or poor reviews can separately expose their submissions to desk rejection.

Why the chairs connect the limits to AI

ICLR is not treating every newcomer paper as AI-generated. The chairs’ narrower argument is that modern systems have lowered the barrier to producing documents that resemble research papers, including work whose claims users may be unable to verify. They say competitive pressure is also encouraging both newcomers and established researchers to generate more papers than the expert community can review.

The chairs report that about 20% of ICLR 2026 submissions had no reciprocal reviewer. Within that group, approximately 15% were sole-authored and around 40% involved authors with multiple no-reviewer submissions. One quarter were desk-rejected, while papers that received full review were accepted at about half the rate of submissions involving program-committee authors.

Those aggregate figures do not establish the quality or provenance of an individual manuscript. The chairs acknowledge that the group contained excellent accepted papers and frame the quota as a way to ration reviewer attention while preserving one route into the conference for teams outside the recognized publication network.

What the experiment has yet to show

The quotas are published conference rules, but their effects will not be measurable until the submission and review cycle progresses. The chairs identify two risks: legitimate interdisciplinary researchers may have publications outside the eligibility list, and established authors may be invited onto papers principally to unlock reviewer eligibility.

ICLR plans to monitor such arrangements and share data from the experiment. The unresolved questions are how many submissions will be flagged, withdrawn or randomly desk-rejected, and whether the limits will reduce review pressure without excluding credible new entrants.

Also read:

• Peter Thiel and the Project of a New Humanity: Immortality, Techno-States, and the Rule of the Chosen"They built a world for everyone. I'm building one for those who survive."
• Who Is A Business Growth Consultant & Why Hire One?
• Scholarcy Does More Than Summarize—But Free Use Stops at One Paper a Day