The growing role of AI in generating scientific literature is raising concerns. Recent studies explore how much of this content is created by AI, revealing a fast-changing landscape that’s hard to fully understand.
Many researchers fear that poor-quality or fake studies made by large language models (LLMs) could flood current systems meant to check for errors. This could harm the trustworthiness of science.
Richard She, a scientist in Singapore, warns it’s an “escalating arms race” between those abusing AI and those trying to stop it. Maria Antoniak, a computer scientist, says the situation is evolving so quickly that experts are unprepared.
In March, an analysis found AI-generated articles likely outnumbered human-written ones online. While AI can help speed up research, it also poses risks like creating fake or low-quality papers.
To tackle this, researchers use tools to detect AI-generated text. However, these tools aren’t perfect—some can’t tell the difference between text edited by AI and fully generated content. They also define “AI-generated” differently, sometimes wrongly flagging human-written work as AI-made.
A study using detection tools on 7,000 journal submissions found a 42% increase in AI-generated content since November 2022. By early 2026, over 30% of peer-review reports were also found to contain AI text.
Other researchers, like She and Antoniak, are trying to track AI-generated papers online. Their efforts are tough due to the massive number of papers published. For example, She analyzed biomedical papers in top journals and found that about 12.5% contained some AI-generated text.
Antoniak’s research on arXiv preprints showed a sharp rise in AI-generated content between 2023 and 2025, especially in computer science.
Experts predict this trend will continue, signaling the start of a new era where AI plays a bigger role in scientific publishing.