Reporting Non-Consensual Intimate Media: An Audit Study of Deepfakes

View on arXiv ← Back to list

Authors: Li Qiwei, Shihui Zhang, Andrew Timothy Kasper, Joshua Ashkinaze, Asia A. Eaton, Sarita Schoenebeck, Eric Gilbert

Published: 2024-09-18 17:01:48+00:00

Comment: under review

AI Summary

This audit study investigates the efficacy of reporting non-consensual intimate media (NCIM) on X (formerly Twitter) using either the platform's non-consensual nudity policy or a copyright infringement claim. Researchers uploaded 50 AI-generated nude images and reported them via these two mechanisms. The study found a stark contrast in effectiveness, with copyright claims leading to rapid content removal while non-consensual nudity reports were entirely ineffective.

Abstract

Non-consensual intimate media (NCIM) inflicts significant harm. Currently, victim-survivors can use two mechanisms to report NCIM - as a non-consensual nudity violation or as copyright infringement. We conducted an audit study of takedown speed of NCIM reported to X (formerly Twitter) of both mechanisms. We uploaded 50 AI-generated nude images and reported half under X's non-consensual nudity reporting mechanism and half under its copyright infringement mechanism. The copyright condition resulted in successful image removal within 25 hours for all images (100% removal rate), while non-consensual nudity reports resulted in no image removal for over three weeks (0% removal rate). We stress the need for targeted legislation to regulate NCIM removal online. We also discuss ethical considerations for auditing NCIM on social platforms.

Key findings

All 25 images reported under the DMCA were successfully removed within 25 hours (100% removal rate), leading to temporary suspensions for the poster accounts. In contrast, none of the 25 images reported under X's non-consensual nudity policy were removed over the three-week observation period (0% removal rate), with no consequences for the poster accounts. This highlights a critical disparity in content moderation efficacy between legally-backed copyright claims and platform-specific privacy policies.

Approach

The researchers conducted an audit study by creating 50 AI-generated nude images across five unique AI-personas and posting them on X using newly created 'poster accounts.' They then reported half the images under X's non-consensual nudity policy and the other half under the Digital Millennium Copyright Act (DMCA) copyright infringement mechanism. The study systematically tracked the speed and success rate of content removal for each reporting method over three weeks.

Datasets

50 AI-generated nude images created by the authors; X (formerly Twitter) platform for the audit.

Model(s)

UNKNOWN

Author countries

USA

← Previous