AI is increasingly writing scientific papers, and it becomes impossible to distinguish them from human-written ones, which is frightening.

AI in Scientific Publishing: Growth and Challenges

What’s Happening
- Volume Increase

According to *Nature*, the number of articles that use AI already exceeds those written solely by humans (Graphite researchers analyzed 55,000 new web pages by the end of March).

- Pros and Cons

AI speeds up the research process, but at the same time a lot of low‑quality work appears.

How AI Participation Is Measured
Study Period Data Volume Key Findings Pangram Labs (7,000 articles + 8,000 reviews) Jan–Feb 2021–Feb 2026 7,000/8,000 From November 2022 growth in works associated with AI is 42 %.
Since the start of 2024, articles where >70 % of the text is AI‑generated have doubled; 30 % of reviews contain AI text. Biomedical journals (Science, Nature, Cell)—~5,000 articles 6 articles are fully written by AI; every eighth article contains elements of AI text. arXiv (computer science) 2020–2025 >124,000 manuscripts Review articles with AI grew from 7 % to 43 %. Unique studies with AI text range from 3 % to 23 %.

Problems and Limitations
1. Reliability of Detectors

- Current tools cannot distinguish “edited” text from fully AI‑generated text.
- False positives are possible—human‑written material may be flagged as AI.

2. Lack of Precise Metrics

There is no universal way to determine the proportion of AI‑created content in the overall scientific literature.

3. Technology Evolution

- As AI advances, new methods for “humanizing” text and evading detectors emerge.
- Solutions such as *watermarks* that explicitly indicate AI involvement have already led to the rejection of 497 conference papers.

Conclusion
The scientific community faces a rapid rise in publications involving artificial intelligence. This opens new opportunities to accelerate research but simultaneously requires the development of reliable detection tools and quality controls to preserve the integrity of scientific literature.