Skip to content

Guide

Desk rejection: the 2026 triggers are in your author list

The rule most likely to end a submission before anyone reads it is no longer about your margins. ICLR 2027 desk-rejects a paper if none of its authors is registered to review, and the author list is fixed from 18 September 2026.

Published 14 August 2026

This one is pre-submission, all of it. Every trigger below is free to fix before the abstract deadline and unappealable at most venues after it. Once a paper is with reviewers the question changes to why papers get rejected

779 ICLR 2026 submissions desk-rejected of 19,525 valid, format-compliant submissions
497 ICML 2026 papers desk-rejected over reviewer LLM policy traced to 398 reciprocal reviewers
11 NeurIPS 2025 papers with a grossly negligent reviewer-author confirmed cases, of 21,575 submissions
18.4% of the NeurIPS 2026 position track, desk-rejected 178 of 969 submissions

If you are aiming at ICLR 2027, read this first

Not in your PDF

Every large venue now desk-rejects over reviewing

Five years ago this table did not exist. Each venue turns a reviewing obligation into a decision about your paper.

Reviewing obligations that cause desk rejection, by venue
Venue and editionWhat it asks of youWhat it desk-rejects
ICLR 2027 (open)One author registered to review 3 papers; anyone on 3 or more papers reviews 6."If none of the authors are registered as a reviewer, it will result in desk rejection for the paper."
ICML 2026One author agrees to review; every author with 4 or more submissions must.Reviewers failing to participate adequately may have "their own submissions desk-rejected".
CVPR 2026Qualified authors review, and "LLMs are NOT allowed to be used to write reviews or meta-reviews".A review deemed 'highly irresponsible' brings "a desk rejection of all papers on which they are an author".
NeurIPS 2026Its Responsible Reviewing Policy for Reviewer-Authors.Desk rejection at the meta-review stage, "a last-resort sanction against grossly negligent reviewer-authors".
ARR, October 2026 cycle (open)Assessed for the whole author team rather than per person."Papers from the teams that do not meet this requirement, and are not covered by an exception, may be desk rejected."
AAAI-27All qualified authors available to review, light load.A violations subcommittee "empowered to impose penalties ranging from desk rejection of papers to multi-year bans".

Quoted from the ICLR 2027 author guidelines, ICML 2026 call, CVPR 2026 reviewer guidelines, NeurIPS 2026 handbook, ARR call and AAAI-27 call, read August 2026. Only the rows marked open have a window now; venues rewrite these between editions.

An all-junior author list can die before the abstract is read

ICLR 2027 wants an accepted publication at one of a closed list of 29 venues, counting primary conference papers only: position papers, tiny papers, blog posts, demos and workshop papers are named as not counting. A lab where nobody has cleared that bar gets one paper each with no eligible reviewer, so its second submission dies.

Your reviews and your paper are held against each other

ICLR 2027 and NeurIPS 2026 withhold your own reviews until you finish the ones assigned to you, and the ICLR review window opens six days after the paper deadline. ICLR's reviewer guidelines add that reviewers who fail to improve a low quality review after an AC warning "may have their own papers desk rejected". Decorative? NeurIPS 2025 reported "the 11 unfortunate cases where one of the co-authors are confirmed to be grossly negligent in their reviews".

Some of it is discretionary, and the venues say so in the text

Read the verbs. CVPR 2026 acts "at the discretion of the PCs", NeurIPS 2026 reserves gross negligence as "a last-resort sanction", ARR says such a team "may be desk rejected", ICLR 2027 alone says "will". A registration you never completed fires on its own; an irresponsible review needs a human to decide.

The real deadline

The abstract deadline is where the decisions get locked

Authors treat the abstract deadline as a formality and the paper deadline as the wall. It is the other way round.

What freezes at the abstract deadline, by venue
What freezesVenue and editionThe rule
Who is on the paper, and who counts as a reviewerICLR 2027, abstract 18 September 2026 AOE"No authors can be added after the abstract submission deadline." Eligibility is computed against the whitelist as it stands then.
Who is on the paperNeurIPS 2026, abstract 4 May 2026 AOE"After this, no changes can be made, except to the author order."
Who is on the paper, then the orderCVPR 2026, abstract 6 November 2025Changes stop at the abstract deadline, reordering at the paper deadline, profiles 16 November.
How many papers each of your co-authors is onICLR 2027: 20. CVPR 2026: 25. AAAI-27: 10. IJCAI-ECAI 2026: 8.IJCAI-ECAI 2026 desk-rejects the excess "identified by submission ID", so which paper dies follows the order they were created.
The abstract text itselfAAAI-27, abstract 21 July 2026The chairs "reserve the right to reject papers without review if they change their abstracts substantially" between the two deadlines. No other venue here claims this.
A complete, current OpenReview profile for every co-authorICLR 2027, NeurIPS 2026, CVPR 2026, ECCV 2026ICLR moderation of profiles without an institutional email "can take up to two weeks". A stale NeurIPS 2026 profile "risks desk rejection". CVPR and ECCV desk-reject "papers with authors who have not completed their profiles".

Dates and rules from the ICLR 2027 call and author guidelines, NeurIPS 2026 call and handbook, CVPR 2026 author guidelines, ECCV 2026 policies, IJCAI-ECAI 2026 call and AAAI-27 submission instructions, read August 2026. Every date but ICLR 2027's has passed; they show the rule's shape.

Seven days between the ICLR deadlines, up to fourteen for a profile

ICLR 2027's deadlines: abstracts 18 September 2026, papers 25 September. A co-author registering with a personal email address goes through moderation ICLR says can take up to two weeks, longer than the gap between those dates. New co-authors need a profile before the abstract goes in, not before the PDF.

How often it happens

The published counts, side by side

Very few venues publish one at all. These did, and they disagree by a factor of five depending on the track.

Published desk-rejection counts by venue and track
Venue, edition and trackDesk rejectedOut ofShare
ICLR 2026, main track77919,525 valid, format-compliant submissionsabout 4.0%
ICML 202693924,661 submissionsabout 3.8%
ICML 2026, reviewer LLM policy violations497the same submissions, traced to 398 reciprocal reviewersabout 2%
NeurIPS 2026, Position Paper Track178969 submissions18.4%
NeurIPS 2026, position track, asked for evidence or rejection123969 submissions12.7%
NeurIPS 2025, main track, reviewer negligence only11 papers21,575 submissionsconfirmed cases
Elsevier journals, figure published April 201530 to 50%articles submitted, before peer reviewan editor's judgement, not a rule

Sources: the ICLR 2026 retrospective; ICML's post on LLM review policy violations, source of the 497; the 939 of 24,661 came from ICML's channels, not a linkable page. The NeurIPS blog on AI-generated papers in the 2026 position track and the 2025 chairs' reflections; and Elsevier Connect, 10 April 2015. Shares are ours where the source gave counts only. The tinted rows are one track at one venue, not a base rate: that screen ran Pangram v3.3.2, flagging any window above 0.75.

The number everyone quotes is from 2015 and describes another decision

Elsevier's figure is that "editors reject between 30 and 50 percent of the articles submitted to Elsevier journals before they even reach the peer-review stage", and that "one of the top reasons for rejection is poor language": an editor judging scope and writing, where a conference applies a rule to a countable property. It is why a better cover letter does nothing at NeurIPS.

Four percent is a floor, and nobody publishes the cause

ICLR 2026 reported 779 desk rejections among "19,525 valid, format-compliant submissions". That denominator has already been screened for compliance, so 4.0% is the rate among papers that passed formatting; the true share is higher by an amount nobody publishes. The 779 is one bucket, never split by cause, so any page naming the commonest trigger has invented it.

Afterwards

Assume there is no appeal in writing

A scored rejection buys you reviews and a rebuttal. A desk rejection buys the next cycle.

Published routes back after a desk rejection, by venue
Venue and editionRoute backWhat it says
NeurIPS 2026, position track, the 178None"A standard desk-rejection that is not subject to appeal under standard circumstances."
NeurIPS 2026, position track, the 123 borderline casesOne tier onlyEvidence of substantial human engagement, by 15 June 2026. The only documented appeal here, closed to most of those it rejected.
ICLR 2026One cause onlyHallucinated-reference rejections came with "an appeal channel to take care of any erroneously flagged papers", built for a false positive, not a right of reply. The word appeal is absent from the ICLR 2027 guidelines.
CVPR 2026, ECCV 2026, ICML 2026, AAAI-27Nothing publishedNo appeal mechanism appears on any author or reviewer page read in August 2026. Not proof none exists, and a bad thing to plan around.

Appeal wording from the NeurIPS 2026 position track post, its 2025 predecessor and the ICLR 2026 retrospective, read August 2026.

Do the co-author check yourself. Hand over the file.

The reviewer registrations and the author list are yours to settle before the abstract deadline, and no manuscript tool can see them. The file is the half someone else can read: specialists go through it against the venue's own guidelines and return the objections with locations, while it is still open.

No manuscript tool, ours included, can see your OpenReview profile, your co-author's reviewer registration, or their submission count. The checklist covers the file half, page limits and anonymity included. Once a paper reaches reviewers, see why ML conference papers get rejected.

phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.