Journal of Online Trust and Safety

Deepfake Abuse and Gendered Digital Creative Violence: Feminist AI Interventions from Mexico

2026-04-20T15:35:54+00:00

Generative AI is transforming image-based sexual abuse and exposing fundamental limitations in existing trust and safety frameworks. While research has focused on the technical details of detection and platform governance, less attention has been paid to how generative AI changes the nature of harm itself. Drawing on an intersectional feminist study of interviews with 12 survivors, legal advocates, activists, technologists, and researchers in Mexico, this article introduces gendered digital creative violence to explain how generative AI weaponizes creative production to generate emotional, reputational, psychological, and structural harm. We show that deepfake abuse extends beyond synthetic content to encompass victim-blaming, institutional failure, evidentiary instability, and affective exhaustion, leaving survivors to shoulder the burden of digital safety. At the same time, feminist organizations have developed AI-enabled counter-infrastructures, most notably the survivor-support chatbot OlimpIA, demonstrating how AI can be redesigned around care, accompaniment, and collective protection. We argue that creative violence offers a transferable framework for understanding emerging forms of generative harm beyond deepfakes by shifting attention from reactive content moderation to the politics of creation itself. This perspective advances trust and safety scholarship by proposing feminist approaches to AI governance grounded in structural prevention, situated ethics, cross-sector collaboration, and survivor-centered design.

Deepfakes in the Global South: Understanding Stakeholder Perceptions, Harms, and Governance Challenges in Sri Lanka, India, and Bangladesh

2026-04-20T15:45:50+00:00

With the rapid proliferation of AI-generated synthetic media across South Asia, communities in Sri Lanka, India, and Bangladesh face deepfake-related harms that existing Trust & Safety frameworks—designed predominantly for Global North contexts—are ill-equipped to address. Through three participatory workshops with 49 stakeholders including journalists, lawyers, fact-checkers, feminist advocates, and civil society practitioners, we investigate how deepfake harms are perceived, experienced, and contested across these distinct yet interconnected contexts. Our findings reveal four cross-cutting patterns: disproportionate gendered targeting of women and marginalized communities; systematic failure of platform moderation for South Asian languages; inaccessible institutional recourse for victims; and strong community preference for locally governed, participatory solutions over externally designed technical fixes. We contribute empirical evidence of deepfake harm in under-researched Global South contexts, a cross-country comparative framework for Trust & Safety analysis, and a set of design and policy implications that center community knowledge, survivor support, and linguistic justice as foundations for a reimagined deepfake governance paradigm.

Evaluating Multilingual Safety Benchmarks for Low-Resource Languages in the Majority World

2026-04-20T17:11:00+00:00

As large language models are deployed across multilingual environments, the benchmarks used to evaluate their safety remain largely designed for high-resource, English-dominant contexts. Trust & Safety systems increasingly rely on automated classifiers and generative models to moderate harmful content across dozens of languages, yet the tools used to assess them often fail to capture linguistic variation and culturally specific harm. This paper presents a structured review of 37 benchmarks relevant to multilingual safety evaluation, 27 of them safety benchmarks, published between 2019 and 2026. Using a nine-dimension taxonomy spanning linguistic authenticity, cultural grounding, and evaluation transparency, we analyze how benchmarks construct evaluation datasets and report safety performance. We identify recurring design patterns: reliance on translated English prompts rather than native-language data, aggregate metrics that obscure cross-language variation, and limited use of locally grounded harm categories or community-informed evaluation. These patterns show that broader language coverage alone does not ensure benchmarks capture how harm is expressed across cultural contexts. We therefore propose a design framework emphasizing native-authored data, disaggregated reporting, locally grounded harm taxonomies, symmetric evaluation of false positives, coverage of dialect and code-switching, and testing under deployment-relevant conditions.

Platform Oversight in Practice: How Language Shapes Procedure, Not Outcome, in Meta’s Oversight Board

2026-04-20T17:12:52+00:00

As private platforms increasingly govern public discourse, legitimacy hinges not only on moderation outcomes but on the institutional processes through which disputes are selected, reviewed, and adjudicated. This study examines the representational structure of platform governance through a case-level analysis of 147 decisions issued by Meta’s Oversight Board between 2020 and 2025, coding geographic distribution, linguistic representation, policy domain, adjudicative outcomes, and procedural pathways. Oversight is globally distributed but internally stratified: representation spans diverse national contexts yet concentrates in a few countries, and non-English cases cluster disproportionately in the Global South. Policy attention favors high-stakes expressive domains. Adjudicative outcomes do not differ significantly by language or geopolitical classification; instead, language shapes procedural routing, as English-language cases are more often resolved through summary reversal and non-English cases proceed disproportionately to full panel review. Institutional design, more than identity characteristics, thus structures how disputes are handled. Because the Board’s archive captures only the subset of disputes entering formal review, these patterns reflect institutional attention and procedural filtering rather than underlying dispute distributions. Legitimacy in corporate digital oversight therefore cannot be assessed through outcome parity or demographic representation alone; case selection and procedural routing are central.

Acceptance and Use of the National Digital Identity System in Indonesia

2026-04-20T15:55:01+00:00

National digital identity systems increasingly enable access to public and private services, but their effective use depends not only on technological infrastructure but also on whether users perceive national digital ID as useful and easy to use. This paper examines the acceptance and actual use of Indonesia’s national digital ID “e-KTP” through an Extended Technology Acceptance Model. Drawing on a nationally representative mixed-methods study of 1,565 households across five major island groups in Indonesia, the study examines how perceived ease of use, perceived usefulness, institutional trust, regulatory awareness, social acceptance, and mandatory adoption shape users’ attitudes, behavioral intention, and actual use of the e-KTP. The findings show that the e-KTP has become a trusted and widely accepted identity because it is familiar, reliable, and consistently recognized across public and private service points. Institutional trust emerges as the strongest factor shaping users’ attitudes toward e-KTP and predictability of acceptance of e-KTP by service providers is driving the actual use. An increase in resident’s awareness of redressal mechanisms and safeguards may increase the use of the e-KTP for private-sector services. The transition from identity coverage to trusted online use depends on institutional credibility, consistent service acceptance, user awareness of safeguards, and robust data governance.

Security without Safety: Queering Cybersecurity in the Age of Digital Transnational Repression

2026-04-20T17:09:27+00:00

Exiled activists increasingly face government-sponsored digital threats, including surveillance, hacking, and spyware, while residing in host states such as the United States, Canada, and the United Kingdom. This phenomenon, known as digital transnational repression (DTR), involves the use of digital technologies to track, intimidate, and silence individuals across borders. Drawing on interviews and questionnaire data with exiled women activists from Global Majority countries, this article focuses on a subset of participants who identify as queer to analyze how intersecting identities of gender and sexual orientation shape their experiences of DTR. We found that exiled queer activists were targeted with spyware and phishing, alongside threats of gendered violence, sexualized harassment and smear campaigns. Situating our analysis within an intersectional feminist framework, we analyze the experiences of exiled queer activists targeted with DTR and identify a critical protection gap that mainstream cybersecurity frameworks and technical-oriented approaches to DTR do not address the associated state-sponsored identity-based attacks and the social, reputational, and psychological harms caused by these threats. We conclude by adopting a duty of care framework, to reconceptualize the defense against DTR, moving beyond digital security toward safety.

Shaping Queerphobia: How X Amplifies Hate Speech against Iranian Queer Activists

2026-04-20T15:33:17+00:00

This article examines how queerphobic hate speech is mobilized against Iranian queer activists on X (formerly Twitter) and situates these dynamics within the heteronormative and authoritarian gender regime of post-revolutionary Iran. While X is framed as a “digital town square” committed to maximal free speech, this study demonstrates that expanded circulation does not distribute vulnerability evenly. Drawing on semi-structured interviews with five Iranian queer activists in the diaspora and a critical technocultural discourse analysis (CTDA) of 59 Persian-language tweets directed at activists, the article analyzes how hate speech operates as a patterned and infrastructurally amplified practice. This paper shows that misgendering, sexualized threats, bodily degradation, and nationalist delegitimation function as public enforcement of gender and sexual norms. Integrating the framework of communicative capitalism, the article argues that X’s engagement-driven logics convert queer visibility into circulation value, intensifying exposure to harassment while weakening moderation protections. These findings position platform-mediated hate as a structural extension of authoritarian gender governance. By foregrounding an underexamined authoritarian context, this study contributes to platformized violence scholarship and calls for more context-sensitive approaches to platform governance in multilingual and politically restrictive environments.

“Our Content Should Be Treated Differently”: Race, Gender, Language, and the Experiences of Quechua Women on Social Media

2026-04-20T17:11:44+00:00

I examine online gender-based violence (GBV) targeted at Quechua women social media users who post in Quechua and how trust and safety systems address this problem. This builds on previous research on how such systems operate in low-resource languages and the literature on online GBV in the Majority World. I use an intersectional framework to examine how the overlapping identities of race, gender, and language shape who has power when posting in a low-resource language (i.e., Quechua). Using a 2024/25 survey of Quechua social media users; interviews with content moderators, linguistic activists, and others; and social media comments about Quechua women influencers on TikTok and YouTube during 2025, I find that Quechua women are targeted with online GBV because of their race, gender, and language. The analysis also suggests that trust and safety systems fail to provide protection for these women, in part because moderation and reporting policies are ineffective and because the AI-based tools used by social media companies perform worse in Quechua. I argue that social media platforms should treat the content of historically oppressed groups differently—with greater protection, community engagement, and investment.

Digital Intersectionality and Marginalization in the Majority World

2026-08-20T17:10:19+00:00

The concepts of ‘intersectionality’ and ‘marginalization’ are empirical necessities in Majority World trust and safety research because the harms that Majority World users face are constituted by multiple systems of power, not by any one axis alone. In this context, marginalization operates at three registers simultaneously: the marginalization of Majority World users by platform infrastructure that are not inherently designed to accommodate their needs; the marginalization of Majority World communities when existing frameworks of harm do not acknowledge or recognize their lived experiences; and the marginalization of Global Majority scholarship by the field’s own institutional and review structures. In response, we initiated this special issue to explore Digital Intersectionality and Marginalization in the Majority World. Here we reflect on our editorial process, from the initial set of 59 submissions we received, to the 8 peer-reviewed papers that we ultimately accepted. We examine these published papers through the lens of intersectionality and marginalization before introducing them across three thematic clusters: moderation systems and AI infrastructure; platform governance; and gendered harm and queer safety. Across these, we corroborate a unifying structural insight: Global North safety frameworks often reproduce structural inequality when deployed in Majority World contexts. We also name an additional dimension of intersectionality that structures the issue itself: the intersection of academic knowledge production and industry practice that defines trust and safety research as a field, and that this journal was designed to sustain. We close with implications for journals, funders, and platforms, and with a reflection on our positionality as editors who are simultaneously Global North-based, Majority World-committed, and situated across the academic-industry divide. This special issue is a call to action for advancing research and practice on trust and safety issues in under-examined regions.