The increasing involvement of artificial intelligence in the creation of scientific articles is becoming a significant concern for researchers. Recent studies indicate that it is now challenging to distinguish between human-written and AI-generated content. A report highlights that by the end of March, the number of AI-influenced articles surpassed those created solely by humans, according to specialists who analyzed 55,000 new web pages.
The community of researchers is alarmed by the implications of this trend, which mirrors broader changes seen across the internet. Tools to detect AI-generated content have been developed, but they often struggle to provide definitive results. Some detection methods cannot differentiate between AI-edited and entirely AI-generated texts, leading to potential misclassifications. For instance, a recent analysis utilizing a tool from Pangram Labs reviewed 7,000 scientific articles and 8,000 reviews submitted to a prominent journal from January 2021 to February 2026. It found a staggering 42% increase in submissions since the emergence of ChatGPT in November 2022, largely attributed to AI participation.
Furthermore, from early 2024 to February 2026, the volume of articles with over 70% AI-generated text more than doubled, and AI-generated content appeared in 30% of the reviews. In a separate study examining 5,000 biomedical science articles published in prestigious journals, it was revealed that six were entirely composed by AI, while one in eight contained some AI-generated text.
Another investigation assessed over 124,000 manuscripts on the arXiv platform and uncovered a notable rise in AI-generated review materials, escalating from 7% in 2023 to 43% in 2025. The percentage of unique research papers incorporating AI text jumped from 3% to 23% during the same period.
A critical aspect of the current situation is the lack of reliable tools to accurately measure the proportion of scientific literature generated by AI. As AI models evolve, so must the detection systems, as individuals may increasingly attempt to "humanize" AI-generated texts to evade detection. One proposed solution includes the introduction of "watermarks" to clearly indicate AI's role in content creation, which recently resulted in the rejection of 497 submissions to a scientific conference.
This rapid shift in the landscape of scientific publishing signifies a transformation that could challenge traditional norms, prompting competitors and the market to adapt to the new realities of AI-generated research.
Informational material. 18+.
