RMIT study finds 7% experienced AI-generated image abuse

The survey covered more than 7,200 adults across the United States, the United Kingdom and Australia.[1] Its 7% figure refers to respondents who had experienced at least one form of AI-generated image-based abuse since turning 18.[1] The study included non-consensual creation or alteration of an intimate image, non-consensual sharing, and threats to share an image.[1] When the definition included all forms of digital image manipulation, 15.1% of respondents said they had been victimised.[3] These are distinct measures, so the broader figure should not be read as the prevalence of the narrower AI-generated abuse category.

Men reported experiencing the abuse at a higher rate than women: 10.2%, compared with 3.8%.[1] The study also reported higher abuse rates among people under 35, people from racial and ethnic minority backgrounds, men and people with disability.[1] People with disability were 1.89 times more likely to report victimisation, while people from racial and ethnic minority backgrounds were 1.70 times more likely to report being targeted.[3] Almost 87% of victim-survivors reported at least one impact.[1] Women reported more negative reactions and impacts, including fear for their safety and effects on mental and physical health.[1]

Text messages and group messages were common sharing channels, accounting for 28.7% and 29.9%, respectively.[3] Among respondents who admitted perpetrating the abuse, former intimate partners were the most commonly reported targets, at 16.7%.[3] The most commonly reported reasons for creating AI-generated intimate images were experimenting with the technology and showing off.[3] Among people who shared such images, the most commonly reported motivation was doing it for fun or as a joke.[1] These reported motives and channels can help teams frame prevention and reporting guidance around the ways the study says abuse occurs.

In a community attitudes study, 88.9% of respondents considered AI-generated image-based sexual abuse totally or somewhat unacceptable.[3] In a vignette experiment, respondents judged creating and distributing AI-generated intimate images of celebrities as less unacceptable than doing the same to an intimate partner.[3] Almost one in five adults reported deliberately viewing AI-generated sexual images of real people, excluding those who viewed them solely to report or remove them.[1] The victimisation study was published in the International Journal of Human–Computer Interaction, and the research program was funded by Google.[1]