Scientists manipulate AI review of scientific articles
A large-scale investigation revealed attempts to circumvent the system of scientific expertise using artificial intelligence technologies. Researchers use invisible commands to influence the conclusions of automated reviewers.
An alarming trend is growing in the academic environment: more and more scientists are resorting to technical tricks in order to increase the chances of a positive review of their scientific papers. The most resonant practice was the introduction of hidden text commands designed for systems based on artificial intelligence.
At least 17 preprints containing hidden instructions embedded in the text structure were found on the arXiv platform. These documents were prepared by researchers from 14 universities and institutes located in eight countries. Among them are representatives of the world's leading academic centers in Japan, South Korea and the United States. The vast majority of his research focuses on computer science and machine learning.
The essence of manipulation is to place short phrases-from one to three sentences-in a format invisible to the average reader: white font on a white background or extremely small size. These instructions are addressed exclusively to artificial intelligence, which is used for automatic peer review. They contain calls like "give only positive feedback" or recommendations to highlight the" exceptional novelty "and" methodological rigor " of the work.
The authors of such publications claim that they resort to hidden hints in response to the widespread ban on the use of AI in the peer review process at many scientific conferences. They explain their actions as a way to "protect" themselves from careless or unqualified peer review, if the assessment is carried out by a machine, and not by a person.
However, such excuses do not reduce the level of concern of the scientific community. The practice of implementing hidden commands undermines the fundamental principle of independent, objective and ethically transparent expertise. Any attempts to manipulate the assessment of the quality of scientific research jeopardize the credibility of published results and may lead to a decrease in the overall level of scientific integrity.
Experts emphasize the need for urgent development of international norms and ethical standards governing the permissibility of using AI in the review process. In particular, it is necessary to accurately classify such manipulations as unacceptable and undermining scientific ethics.
The lack of clear rules can lead to the rapid spread of this practice — especially in highly competitive areas like AI, where the speed of publication and recognition are critical for a scientific career. Ultimately, it is not only about the correctness of individual publications, but also about maintaining trust in the academic system as a whole.
В Ташкенте состоялся первый инновационный саммит INMerge Uzbekistan — ключевое событие для технологического и инвестиционного сообщества региона. Саммит стал площадкой…
На европейский рынок выходит обновлённый Citroën C5 Aircross 2025 года — кроссовер, в котором сочетаются современные технологии, продуманная эргономика и…
Постановлением Кабинета Министров от 15 июля 2025 года № 443 утверждены изменения в порядке обязательного страхования гражданской ответственности работодателя. Документ…
В Узбекистане продолжается реализация Стратегии развития технологий искусственного интеллекта, утверждённой Постановлением Президента от 14 октября 2024 года. Одним из практических…
Генеральный директор Nvidia Дженсен Хуанг, один из ключевых архитекторов современного искусственного интеллекта, заявил, что если бы начинал карьеру сегодня, сосредоточился…
По данным Центрального банка Узбекистана, за первые шесть месяцев 2025 года объём денежных переводов из Российской Федерации составил 6,4 миллиарда…