← Retour au blog
tech 20 July 2026

Over 30% of New ArXiv Submissions Now Read as AI-Written

A recent study shows that a third of new submissions on ArXiv appear AI-written. Let's explore the implications and challenges of this growing trend.

Article inspired by the original source
Over 30% of new ArXiv submissions now read as AI-written ↗ unslop.run

Introduction

ArXiv, the widely used online repository for scientific articles, has recently revealed a surprising trend: over 30% of its new submissions appear to be AI-written. This statistic, derived from a comprehensive study, raises exciting and complex questions about the future of academic and scientific research.

Study Methodology

The study analyzed the full text of 12,750 papers submitted to ArXiv between 2021 and 2026. A sophisticated detector, calibrated specifically for academic writing, was used to assess each paper. This detector was designed to minimize false positives, with a threshold set so that only 0.4% of pre-ChatGPT papers were marked as AI-written. This ensured that the results accurately reflect the influence of modern AI systems.

Key Findings

The results show a dramatic increase in the number of submissions marked as AI-written, rising from 0.4% in 2021 to about 32% in the last complete quarter of the study. This rise occurred in two main waves, peaking at nearly 39% in early 2026. Fields such as computer science and quantitative biology saw significant increases, with 65% and 56.3% of recent submissions marked as potentially AI-written, respectively.

Implications for Scientific Research

This trend has profound implications for the scientific community. On one hand, it could accelerate the production and dissemination of knowledge. On the other hand, it raises challenges regarding credibility and academic integrity. Researchers now have to navigate a landscape where distinguishing between human and AI-generated writing becomes increasingly difficult.

Challenges and Potential Solutions

A major challenge is ensuring that AI-generated writing adheres to academic standards. Institutions might develop guidelines and training to help researchers use these tools ethically. Moreover, improving the transparency of the algorithms used could enhance trust in the results presented.

Conclusion

The impact of AI on academic writing is undeniable and rapidly evolving. This study highlights the need for continuous adaptation by academic communities to integrate these technologies responsibly. Ultimately, collaboration between humans and AI could unlock new perspectives for scientific research.

Let's discuss your project in 15 minutes.

IA ArXiv rédaction académique détection de l'IA recherche scientifique
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call