REDDIT · REDDIT// 3d agoNEWS

ICML 2026 reviewers triggered by AI traps

A researcher at ICML 2026 has reported an unprofessional reviewer who issued "fake references" and "personal insults" after falling for a hidden prompt injection trap embedded in the submission. This incident is part of a broader controversy at the conference, where efforts to catch reviewers using LLMs have resulted in a breakdown of professional discourse and technical accuracy.

// ANALYSIS

The ICML 2026 review cycle has become an adversarial battlefield between anti-AI policies and hallucinating reviewers.

–Hidden "trap" text designed to catch LLM-using reviewers has backfired, leading reviewers to accuse authors of "unethical behavior" and "integrity concerns" when the trap is triggered.
–Authors report receiving "mathematically nonsensical" proofs and citations to non-existent or irrelevant papers, strongly suggesting reviewers are offloading their evaluations to AI.
–The split between Policy A (no AI) and Policy B (limited AI) tracks has created a "competitiveness gap," where authors in the non-AI track feel penalized by harsher, more pedantic reviews.
–Program Chairs are being overwhelmed with appeals as the traditional peer-review trust model collapses under the weight of generative AI friction.

// TAGS

icml-2026researchai-ethicspeer-reviewllmpolicy-regulationicml-2026-peer-review

DISCOVERED

3d ago

2026-04-08

PUBLISHED

3d ago

2026-04-08

RELEVANCE

8/ 10

AUTHOR

Martinetin_