ICML Takes Strong Stance Against AI Misuse in Peer Review

The 2026 International Conference on Machine Learning (ICML), scheduled for July in Seoul, has rejected 497 papers—approximately 2% of total submissions—due to violations of its artificial intelligence (AI) use policies in the peer review process. This decision stems from the conference's reciprocal review policy, which mandates that authors must also review other submitted papers. The ICML organizers implemented a watermarking system to detect the use of large language models (LLMs) in peer reviews, leading to the rejection of papers where AI-generated reviews were identified.

Nature

The watermarking system embedded hidden instructions in the research papers distributed for review. If an author utilized an LLM to generate their review, these instructions prompted the model to include specific phrases that indicated AI involvement. The organizers stated, “We hope that by taking strong action against violations of agreed-upon policy we will remind the community that as our field changes rapidly the thing we must protect most actively is our trust in each other.”

Nature

The ICML's actions received mixed reactions within the research community. Many researchers expressed support on social media, suggesting that other conferences could adopt similar measures. Some advocated for stricter penalties, such as banning rejected authors from resubmitting. However, dissenting voices emerged, including Zhengzhong Tu, a computer scientist at Texas A&M University, who argued that the policy might discourage reviewers and lead to the use of LLMs for producing "meaningless reviews."

Nature

The ICML's decision highlights a growing concern regarding the use of AI in academic peer review. A 2025 survey from Frontiers indicated that over half of researchers have employed AI in their reviews, often in violation of established guidelines. This situation underscores the need for clearer policies on responsible AI use within the academic community. In response to these challenges, the ICML has introduced two peer-review streams for the first time: one permitting limited LLM use and another strictly prohibiting it, allowing authors and reviewers to select their preferred approach.

Nature

Marie Soulière, head of editorial ethics and quality assurance at Frontiers, commented on the ICML's actions, stating that the case illustrates a research community in need of clear guidance on responsible AI use, particularly in peer review. The ICML organizers emphasized the importance of maintaining trust within the community as AI technologies evolve.

Nature

While the ICML's rejection of papers is a significant step, there is ongoing debate about the effectiveness of such policies. Critics like Zhengzhong Tu raise concerns that banning AI use may lead to unintended consequences, such as discouraging reviewers from participating altogether. The discussion around the role of AI in peer review continues to evolve, with no consensus on the best path forward.

Nature

• “We hope that by taking strong action against violations of agreed-upon policy we will remind the community that as our field changes rapidly the thing we must protect most actively is our trust in each other,” — ICML Organizers

Nature

• “It will only demotivate all the reviewers,” — Zhengzhong Tu, Computer Scientist, Texas A&M University

Nature

• “What the ICML case shows is a research community in need of clear guidance on responsible AI use, including use in peer review,” — Marie Soulière, Head of Editorial Ethics, Frontiers

Nature