2% of ICML papers desk rejected because the authors used LLM in their reviews

ICML 2026 desk-rejected 497 papers (~2% of submissions) because their reciprocal reviewers violated the agreed-upon no-LLM policy (Policy A) during peer review. 795 reviews by 506 reviewers were flagged via a custom detection method (not generic AI-text detectors) and manually verified; violating reviews were removed and repeat offenders were expelled from the reviewer pool. The conference operates a two-policy framework—conservative (no LLM) and permissive (LLM allowed for limited use)—reflecting divided community opinion on AI in peer review.