"Violation of my body:" Perceptions of AI-generated non-consensual (intimate) imagery

TL;DR

This study investigates public attitudes toward AI-generated non-consensual intimate imagery ($AIG-NCII$) using a vignette-based survey (n = 315 U.S. participants) with a $2 \times 3 \times 3$ factorial design across creator, content, and intent, analyzed via cumulative link mixed models ($CLMM$) and qualitative coding. It finds that creation and sharing of AIG-NCII are generally viewed as unacceptable, while seeking out such content is comparatively more acceptable, with contextual factors (creator identity, intent, and action) and individual factors (sexual consent attitudes and gender) shaping these judgments. Positive attitudes toward establishing consent ($SCS-R2$) and concerns about genuine intimate imagery ($GII$) and non-consensual distribution ($NDII$) correlate with lower acceptability of non-consensual AI imagery. The results highlight policy and design implications, advocating for consent-centric norms, deterrence strategies in information access, and a representational privacy lens to address the evolving risks of IBSA in AI-enabled media.

Abstract

AI technology has enabled the creation of deepfakes: hyper-realistic synthetic media. We surveyed 315 individuals in the U.S. on their views regarding the hypothetical non-consensual creation of deepfakes depicting them, including deepfakes portraying sexual acts. Respondents indicated strong opposition to creating and, even more so, sharing non-consensually created synthetic content, especially if that content depicts a sexual act. However, seeking out such content appeared more acceptable to some respondents. Attitudes around acceptability varied further based on the hypothetical creator's relationship to the participant, the respondent's gender and their attitudes towards sexual consent. This study provides initial insight into public perspectives of a growing threat and highlights the need for further research to inform social norms as well as ongoing policy conversations and technical developments in generative AI.

Figures (7)

• Figure 1: Respondents' perceptions of acceptability across all vignettes; each vignette is defined by the creator / action / intent. Vignettes are grouped by action and ordered (from bottom to top) by increasing unacceptability of creation.
• Figure 2: Heatmap of acceptability for creation and seeking_out when the action is performing a sexual act.
• Figure 3: Heatmaps comparing acceptability of creation and seeking_out for AIG-NCII to similar actions for GII also created in an intimate relationship. See Appendix \ref{['ap:GIIHeatmaps-full']}, Figure \ref{['fig:GIIHeatmaps-full']} for heatmaps including all forms of sharing.
• Figure 4: Odds ratios and confidence intervals for the controlled and uncontrolled IVs in the models for creation, private_sharing, public_sharing, resharing, and seeking_out. Statistically significant IVs ($p<0.05$) are in bold with an asterisk. Y-axis is on a logarithmic scale and trims are indicated by red triangles.
• Figure 5: Heatmap of acceptability for creation and seeking_out when the action is performing a sexual act. Darker indicates more respondents.