ICML Detects Reviewers Using LLMs, Rejects Papers

Organizers for the International Conference on Machine Learning (ICML) told Nature they rejected 497 papers — about 2% of submissions — after detecting authors had used large language models to review others' work. Conference staff embedded hidden prompts in submissions that exposed LLM-generated review text, identifying 506 reviewers and 795 suspect reviews. The move aims to enforce peer-review rules and prompt clearer conference policies.