Two papers including fake authors reported to have been accepted as Oral

Recently, at major machine learning conferences such as NeurIPS, WACV, ECCV, serious cases of LLM-generated fabricated citations, fake author lists, and LLM slop have been discovered during the review process.

A significant number of the papers reviewers examined were confirmed to be cases where LLM outputs were submitted as-is, including misuse of technical jargon and generation of meaningless sentences. This problem stood out particularly at NeurIPS and WACV, and there were even cases where papers containing false information passed expert review and were accepted as Oral.

This problem appears not only in paper submissions but also in the review process itself. About 21% of ICLR 2026 reviews were AI-generated, and the proportion of reviewers using LLMs to write reviews is also increasing rapidly. In addition, attack methods that adversarially rewrite paper abstracts to manipulate AI reviewer scores have also been confirmed, threatening the reliability of the academic ecosystem.