[KongchangAI]
· 2 min read· 1,073 words

ICML 2027's New Submission Rules: Can Two-Stage Review and Reciprocal Reviewers Curb the Submission Explosion?

ICML 2027's New Submission Rules: Can Two-Stage Review and Reciprocal Reviewers Curb the Submission Explosion?

ICML 2027 uses two-stage review and reciprocal reviewers to control submissions and reviewer load; community reaction is positive but concerns about misjudgment remain.

ICML 2027's call for papers sparked discussion in the academic community. Facing runaway submission volume, the conference did not simply expand the reviewer pool but instead tackled both process and responsibility matching: on one hand, it introduced two-stage review, with ACs pre-screening and SACs verifying first-stage rejections, using low-cost human filtering to block obviously substandard submissions; on the other hand, it implemented a reciprocal reviewer system, requiring each submission to designate one reviewer (assigned to at most 6 papers), mandating that authors listed on 3 or more papers join the Program Committee, capping individual submissions at 20, and allowing each author at most one paper without a qualified reciprocal reviewer. Overall community feedback is positive, viewing the link between authorship count and reviewing obligations as fair, but the risk of misjudgment in rapid first-stage rejection remains the main concern; some also question whether the rules may encourage teams to split submissions or commodify reviewer identity.

Two-Stage Review: ACs Screen First, SACs Verify

The core of the new process is splitting review into two stages. After a paper is submitted, the Area Chair (AC) first determines whether it enters the full review process. If the AC proposes a first-stage rejection, that decision must be confirmed by a Senior Area Chair (SAC). Authors will receive a brief explanation of the reasoning by February 23.

The intent behind this design is clear: use low-cost human pre-screening to filter out submissions that are obviously below the bar, preventing a large number of low-quality papers from consuming full review resources. SAC verification serves as a corrective mechanism to reduce the probability of wrongful rejection caused by a single AC's rapid judgment. However, as some pointed out in Reddit discussions, ACs reading a large number of papers in a short time may still miss valuable work—SAC checks can mitigate but not eliminate this problem.

An Area Chair (AC) is a mid-level role in the top conference review system, responsible for coordinating the review process for several papers, synthesizing review comments, and providing a meta-review. A Senior Area Chair (SAC) is typically a more senior researcher whose role is to oversee the work of several ACs, handle disputes, and safeguard quality. In the traditional process, the AC's meta-review essentially determines a paper's fate, but all papers go through at least a full external review; the change at ICML 2027 is to move the AC's judgment ahead of external review, giving ACs critical "gatekeeping" power, while SACs shift from quality supervisors to corrective reviewers. This shift in the center of gravity of roles is key to understanding the change in responsibilities under two-stage review.

Reciprocal Reviewers: Linking Authorship to Reviewing Obligations

More controversial than two-stage review is the "reciprocal reviewer" system. Each submission must designate one reviewer from its author list, and each reviewer can be assigned to at most 6 papers. At the same time, any author listed on 3 or more papers must join the Program Committee. Papers that do not comply may be desk rejected.

This set of rules directly links submission volume to reviewing contribution, and the logic is quite straightforward: the more you submit, the more you need to take on reviewing work for the community. The conference has also set a stricter restriction—each author may have at most one paper without a qualified reciprocal reviewer. In other words, if your co-authors are people who cannot review (undergraduates, interns, people outside the field), you have only one "exemption" opportunity, after which you will be blocked.

The practical impact of this rule may be greater than it appears. It forces researchers to consider reviewer capacity allocation in advance when organizing author lists, rather than fully externalizing reviewing responsibility. For teams with many industry authors or interdisciplinary collaborators, this may mean the need to plan submission strategies more carefully.

The Program Committee (PC) is the formal set of conference reviewers, usually comprising thousands of researchers, responsible for assigning and reviewing submissions by area. Joining the PC means committing to complete a certain number of review tasks during the review cycle and to comply with the conference's deadlines and confidentiality requirements. In the past, PC members mostly volunteered or were invited by ACs, with no mandatory connection to submission behavior; ICML 2027's mandatory inclusion of authors listed on 3 or more papers into the PC essentially turns PC membership from "voluntary service" into an "obligation triggered by submission volume." This changes the logic of PC composition and also means that more authors need to understand the review system's operations and review standards in advance.

Submission Caps and Mandatory PC Participation

In addition to the reciprocal reviewer system, the new rules set several other gates: no individual may appear on more than 20 submissions; authors listed on 3 or more papers must serve on the Program Committee. Together, these rules form a "submission volume—review volume" matching framework that attempts to control reviewer load at the source.

Discussions on Reddit generally agree that runaway reviewer burden has persisted for years and that tying authorship count to reviewing obligations is fair. But some also raise questions: will such rules simply push the problem elsewhere—for example, encouraging teams to split submissions, or turning reviewer identity into a tradable resource?

The Risk of Misjudgment in Rapid Pre-Screening

The most concentrated concern in community discussions remains first-stage rejection. The quality of an AC's rapid-reading judgment depends on the AC's experience, field match, and the submission volume in that cycle. If submission volume continues to grow, the time ACs can spend on each paper will be further compressed, and the probability of misjudgment will rise accordingly. Although SAC verification provides an additional safeguard, SACs also face time constraints, and the depth of verification may be limited.

From a conference governance perspective, two-stage review is an attempt to trade process for efficiency. Whether it works depends heavily on whether AC and SAC assignments are reasonable and whether the criteria for first-stage rejection are clear enough. If the criteria are vague, authors may still find it difficult to understand why they were rejected after receiving the explanation, and disputes will shift to the appeals process.

Overall, the direction of ICML 2027's combination of measures is reasonable, but its effectiveness needs to be tested in practice. The reciprocal reviewer system is the most structural change among them, as it changes the responsibility relationship between authors and reviewers rather than merely adjusting the review process. For security researchers who frequently submit to top conferences, planning the reviewer configuration in author lists in advance may be more important than ever.

Desk reject means a paper is directly rejected by the Program Chairs or an AC before entering the full review process, usually for submissions that violate formatting, anonymity rules, page limits, or are clearly outside the conference's scope. Traditionally, desk reject criteria are relatively objective and mechanically determinable; while ICML 2027's first-stage rejection is made by ACs based on judgments of paper quality, though still called "rejection," it is closer in nature to a rapid quality screening. The key difference between the two is that the former is less controversial and has limited room for appeal, while the latter involves substantive judgment of research value, and therefore requires higher clarity in criteria and higher quality in reasoning explanations, and is also more likely to trigger authors' doubts about misjudgment.

#ICML 2027#two-stage review#reciprocal reviewer#submission explosion#peer review reform
Share:

Related articles