Area Chair Proposes Desk-Rejecting ~50% of Submissions, Sparking Debate

In early October, a researcher serving as an Area Chair for ICML/ICLR/NeurIPS put forward a controversial suggestion: ACs should be allowed to desk-reject roughly 50% of the papers in their batch outright. His reasoning was that ACs can quickly scan all papers in their batch, identify submissions with virtually no chance of acceptance, and avoid wasting reviewers' time. After being shared by multiple scholars, the proposal triggered widespread discussion.

Confirmed

Core of the proposal: ACs could directly desk-reject about 50% of submissions in their batch, aiming to cut the time reviewers spend on papers with zero chance of acceptance.
One context for the debate is the academic community's frustration over the flood of AI-generated low-quality papers (AI slop); in the shared threads, several scholars voiced support for stricter pre-screening to combat such submissions.

Not Yet Confirmed

The proposal remains a personal suggestion; there is no evidence that ICML, ICLR, or NeurIPS has officially adopted or piloted such a mechanism.
The "50%" figure comes from the proposer's personal judgment and lacks systematic data support.

Why It Matters

Reviewing workload is a core pain point for top ML conferences today; ever-expanding submission volumes are straining review quality, and how pre-screening power is allocated will directly determine which papers live or die.
The HCI community has already pushed back: Pedro Lopes of UIC, responding to new review arrangements for conferences such as CHI 2026, argued that whether desk rejects should be aggressively enforced depends on the case—if an AC reviews only a small pile of papers in the same area (say 10 papers), forcing a desk-reject quota could wrongly kill papers that would otherwise be accepted. This suggests that "cutting half in bulk" may work in large batches but carries higher risks in small AC piles.

2026-10-04 ~ 2026-10-05 · 6 related posts

Primary sources

Area Chair proposes letting ACs desk reject ~50% of papers at ICML/ICLR/NeurIPS — maksym_andr · 2026-10-04 17:50 (UTC+08:00)

[source] Area Chair proposes letting ACs desk reject ~50% of papers at ICML/ICLR/NeurIPS — maksym_andr · 2026-10-04
ICML area chairs push to desk-reject ~50% of submissions amid wave of AI slop papers — HildeKuehne · 2026-10-05
CHI Review Debate: Pedro Lopes Argues Desk-Rejecting Half a Small AC Pile Can Backfire — plopesresearch · 2026-10-05
NeurIPS ACs debate desk-reject quotas: small paper piles make quotas statistically unfair — plopesresearch · 2026-10-05

2 near-duplicate retellings: AmartyaSanyal · menhguin