X (formerly Twitter) takes swift action when taking down deepfake nude images that are reported as copyright violations — but not when they’re reported under “nonconsensual nudity,” a study has found.
The paper, published by researchers at the University of Michigan and Florida International University, is an audit of X’s reporting systems and hasn’t yet been peer-reviewed, 404 Media reported. Researchers created five AI “personas” of young white women (to prevent further variables of race, gender, and age) and then made 10 replica images of each, resulting in 50 images. In terms of the ethics around generating deepfake porn themselves, researchers said these images underwent a “rigorous verification process” to ensure they didn’t represent an existing individual.
They posted these images to X on 10 “poster accounts” they created, and then they created five X accounts to report the images. Twenty-five images were reported as Digital Millennium Copyright Act (DMCA) violations, and the other 25 were reported as …