A Large Scale Social Web Audit of AI Generated Text Detection Systems

Authors

  • Arka Dutta Rochester Institute of Technology
  • Utkarshani Jaimini University of Michigan, Dearborn
  • Utkarsh Bhatt Indian Institute of Technology Kharagpur
  • Sara Shree Muthuselvam University of South Carolina
  • Amitava Das BITS Pilani Goa
  • Ashiqur Rahman KhudaBukhsh Rochester Institute of Technology

DOI:

https://doi.org/10.1609/icwsm.v20i1.42660

Abstract

This paper makes three contributions. First, we exploit temporal signals to conduct an in-the-wild audit of a broad suite of AI-generated text detection (AGTD) systems. Our in-the--wild audit reveals that state-of-the-art (SoTA) AGTD systems exhibit considerable false positives. Second, our audit demonstrates that AGTD systems disfavor liberal political discourse and flags them more often as AI-generated as compared to conservative political discourse. Finally, we extend anticontent sampling approach to robustify existing AGTD systems.
ICWSM-2026 Proceedings Cover

Downloads

Published

2026-05-25

How to Cite

Dutta, A., Jaimini, U., Bhatt, U., Muthuselvam, S. S., Das, A., & KhudaBukhsh, A. R. (2026). A Large Scale Social Web Audit of AI Generated Text Detection Systems. Proceedings of the International AAAI Conference on Web and Social Media, 20(1), 677–690. https://doi.org/10.1609/icwsm.v20i1.42660

Issue

Vol. 20 No. 1: Proceedings of the Twentieth International AAAI Conference on Web and Social Media

Section

Full Papers