🤖One-Third of New arXiv Papers Machine-Written
AI is rewriting academic research at scale
TL;DR
A new study finds that about one-third of recent papers on arXiv are machine-written. Computer science leads with a flagged share of 65%, while mathematics remains untouched at just 0.7%. This shift raises questions about the future of academic research.
About one-third of new arXiv papers read as machine-written, according to a recent study that scored 12,750 full-text papers. The detector flags genuine human writing at only 0.4%, with a false-positive rate calibrated for academic text. This shift is most pronounced in computer science, where the flagged share hits 65%. If you're working on AI or machine learning research, this trend directly impacts your field's credibility and peer review processes. The study sampled papers from January 2023 to July 2026, with a peak flagged share of nearly 40% in early 2026.
Key Points
The study scored full text of 12,750 arXiv papers from January 2023 to July 2026.
Computer science leads with a flagged share of 65%, while mathematics is at just 0.7%.
Detector flags genuine human writing at only 0.4%, clearing 99.6% of pre-LLM text.
The peak flagged share nears 39% in early 2026, up from a flat rate of 0.4% through 2021 and 2022.
Each field's control sample size is 200 papers, with only eight false positives across all fields.
Why It Matters
If you're working on AI or machine learning research, the rise of machine-written papers impacts your credibility and peer review processes. Computer science leads with a flagged share of 65%, while mathematics remains untouched at just 0.7%. This shift raises questions about academic integrity and how we verify originality in scientific writing.
Frequently Asked Questions
Why does this matter?
If you're working on AI or machine learning research, the rise of machine-written papers impacts your credibility and peer review processes. Computer science leads with a flagged share of 65%, while mathematics remains untouched at just 0.7%. This shift raises questions about academic integrity and how we verify originality in scientific writing.
What happened?
A new study finds that about one-third of recent papers on arXiv are machine-written. Computer science leads with a flagged share of 65%, while mathematics remains untouched at just 0.7%. This shift raises questions about the future of academic research.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.