AI-generated image-based sexual abuse: A scoping review

This is the authors' abstract. We don't add key points for this paper.

AI-generated image-based sexual abuse, including non-consensual deepfake pornography, nudification, synthetic intimate imagery, and AI-generated child sexual abuse material, has become an increasingly visible form of technology-facilitated sexual violence. The evidence base, however, remains dispersed across empirical, legal, ethical, child-protection, and platform-governance scholarship and marked by conceptual inconsistency. Unlike prior reviews centered on particular harms, populations, or evidence types, this scoping review integrates these interdisciplinary literatures to examine how AI-generated image-based sexual abuse is defined and conceptualized and to identify the main areas of inquiry represented in existing scholarship. Following PRISMA-ScR guidance, systematic searches were conducted in Web of Science, PubMed, Scopus, and PsycINFO. Twenty-eight studies met the eligibility criteria. Data were charted and synthesized narratively using thematic analysis. Across studies involving identifiable persons, three recurring definitional features were evident: non-consent, AI mediation, and sexualized representation, although terminology remained contested. AI-generated child sexual abuse material was treated as a distinct boundary category: material involving an identifiable child overlaps with this person-centered AI-IBSA frame, whereas wholly synthetic material without an identifiable child does not. Four main areas of inquiry were identified: nature and extent; harms and lived experiences; public perceptions and social meaning; and responses, regulation, and prevention. The literature documents gendered patterns of abuse and reported psychological, relational, social, and institutional harms, while legal and regulatory responses remain fragmented and under-evaluated. Major gaps remain in prevalence measurement, longitudinal research, intersectional analysis, and intervention evaluation. The review’s principal contribution is an integrated cross-domain account of how definitions and measurement shape which behaviors, populations, and harms become visible and how institutions respond.