AI BriefingOver 30% of New arXiv Papers Identified as AI-Written
Key point
About 32% of newly submitted arXiv papers were identified as containing AI-written text, with computer science showing the highest rate.
Details
An analysis of the full text of 12,750 arXiv papers found that about 32% of papers were identified as containing AI-written text in the most recent quarter, with the figure peaking at 39% in early 2026.
The results varied significantly by field:
- Computer Science: 65.0% (highest rate)
- Quantitative Biology: 56.3%
- Electrical Engineering: 51.3%
- Economics & Finance: 47.0%
- Mathematics: 0.7% (lowest rate)
The study was designed to minimize false positives from the detection tool by using papers from 2021-2022, before the emergence of ChatGPT, as a control group. The AI detection rate for the control group remained at around 0.4%, but the rate of AI-written content rose sharply across all fields following the release of ChatGPT.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Toss Doctor V2 Released: Maximizing QA Efficiency with LLM-based Automated Testing and Self-healing