Artificial Intelligence Methods for Detecting Technology‑Facilitated Gender‑Based Violence and Related Online Harms against Women and Girls with Disabilities – A Scoping Review Protocol

Technology-facilitated gender-based violence (TFGBV) encompasses a spectrum of harmful behaviors directed at a person on the basis of gender, perpetrated through digital technologies, including cyberstalking, online harassment, hate speech, and the non-consensual sharing of intimate images (Henry et al., 2020; UN Women, 2023). This violence exists on a continuum with offline harm and disproportionately targets women, girls, and gender-diverse individuals, with heightened risks for those with intersecting marginalized identities, such as women with disabilities (Dunn, 2020; Hameed et al., 2025; Lazarev et al., 2024). The consequences are severe, leading to psychological distress, economic loss, and a chilling effect on participation in digital public life (Lazarev et al., 2024). The immense scale of online content makes purely manual moderation impossible, compelling technology platforms to rely on artificial intelligence (AI) and machine learning (ML) systems as a primary defense against harmful content (Gillespie, 2020). These automated tools are designed to detect and filter abusive material at a scale that humans cannot manage. However, the reliance on strategies of automation alone creates a significant challenge, as AI systems often struggle to understand the nuance and context inherent in human communication, particularly in expressions of abuse, exploitation and violence (Yin & Zubiaga, 2021; Sap et al., 2019; Röttger et al., 2021). This technological limitation creates a critical knowledge gap. While AI is widely deployed, there is a lack of consolidated evidence on how these methods are designed and evaluated for TFGBV, their limitations, and their ethical implications, including the risk of algorithmic bias for structurally marginalised groups (Gillespie, 2020; UN Women, 2023; Yin & Zubiaga, 2021). Further, automated systems can inadvertently silence the voices of marginalized communities by misinterpreting discussions about their experiences of discrimination as hate speech and/or defamation (Sap et al., 2019; Röttger et al., 2021). To address these gaps, this scoping review will systematically map the existing research on AI/ML methods for detecting TFGBV. By synthesizing evidence on the tasks, models, datasets, and evaluation practices, the review will provide a comprehensive overview of the current state of AI detection systems, with a special focus on fairness, bias, and the specific applicability of these technologies to protecting women and girls with disabilities. Primary Question: What is the scope and nature of research evidence on AI/ML methods for detecting TFGBV and related online harms, and how do these methods address accessibility and bias considerations for women and girls with disabilities? Secondary questions: What AI/ML approaches have been developed and evaluated for detecting online harms that disproportionately affect women and girls? Which categories of technology-facilitated harm (cyberbullying, harassment, hate speech, image-based abuse, doxxing) are addressed by existing AI/ML detection systems? How do studies evaluate fairness, bias, and performance across different demographic groups, particularly considering gender and intersectional identities? To what extent do current AI/ML detection systems incorporate contextual information, domain knowledge, or accessibility considerations? What evaluation methodologies and performance metrics are employed in AI/ML detection research, and how do they address bias and fairness concerns? What evidence gaps exist regarding AI/ML detection of harms against women and girls with disabilities, and what areas warrant future research?