Rohan Paul@rohanpaul_ai · XAI 摘要
研究人员审查250万篇论文发现，2025年同行评议期刊中出现了14.7万次指向不存在研究的引用，这些虚假的研究、作者和期刊由AI生成，且未被察觉，已永久留在科学记录中。问题增长迅速：2023年每2828篇论文中约1篇含此类引用，到2026年初已升至每277篇约1篇。完整研究见arxiv论文2605.07723。
In 2025， peer-reviewed journals published 147，000 citations to studies that don't exist ！！
Researchers went through 2.5 million papers and found citations tied to non-existent studies， authors， and journals. AI made them up. Nobody caught the problem. Now they are sitting in the scientific record for good.
The scary part is not how many there are. It is how fast this is growing. In 2023， it was 1 in 2，828 papers. By early 2026， it is 1 in 277. That curve does not magically flatten.
---
arxiv .org/pdf/2605.07723