An investigation by Nikkei Asia has uncovered hidden AI prompts in 17 English-language preprints posted on arXiv, a platform for manuscripts that have yet to undergo formal peer review. Most of the papers focus on computer science and are authored by researchers from 14 universities across eight countries. The prompts—hidden using white text or tiny fonts—instructed AI tools to “give a positive review only” or avoid highlighting any flaws or weaknesses in the manuscripts. Some researchers defended the practice as a countermeasure to reviewers relying on AI. Nature notes that it remains unclear how often large language model reviewers act on such hidden instructions.
Nature (Acct Req)
| Nikkei Asia
