Tipping Point: Online Violence Against Women in the Age of AI

UN Women’s commissioned report Tipping Point: Online Violence Impacts, Manifestations and Redress in the AI Age approaches online violence as a question of whether women can continue to participate in public life. The cross-country survey, fielded in late 2025 in five languages, analysed responses from more than 640 women human-rights defenders, activists, journalists, media workers and other public communicators in 119 countries. The report says 70% of respondents had experienced online violence in the course of their work, 24% believed they had experienced AI-assisted online violence, and 41% linked offline attacks, abuse or harassment to online violence. These are self-reported findings among people in particular public-facing roles, not prevalence estimates for women worldwide.

The study’s point is not to attribute every new form of harm to artificial intelligence, but to ask how AI can accelerate, scale or alter existing attacks. It discusses the non-consensual sharing of sexualized images, deepfakes or manipulated imagery, online sexual harassment and other technology-facilitated harms; it also documents how threats online can extend into offline safety. Twenty-four per cent of the overall respondents reported AI-assisted online violence; among women journalists and media workers, the figure was 19%. These findings reflect respondents’ understanding of their own experiences. They do not mean that the use of generative AI can be independently confirmed in every incident, and the focus on deepfakes should not obscure stalking, threats, doxxing and coordinated harassment that cause harm without AI.

Tipping Point builds on The Chilling, the 2020 UNESCO–International Center for Journalists study, making some comparisons possible. Among women journalists and media workers, the share who said they self-censored on social media because of online violence rose from 30% in the 2020 survey to 45% in 2025. The proportion linking offline harm to online attacks rose from 20% to 42%. These comparisons are informative, but the respondents, group composition and survey timing were not identical; the differences should not be presented as a precisely measured trend caused by a single factor. A more careful conclusion is that both research rounds connect digital attacks with withdrawal from public expression, psychological harm and offline risk.

The report also finds that seeking redress is not absent. In the 2025 survey, 22% of women journalists reported incidents to police and 14% took legal action; both shares were higher than the 11% and 8% reported in the 2020 study. Yet more reporting does not mean effective remedies are in place: the research still records police reluctance to investigate, victim-blaming and gaps in legal protection. Analysis therefore cannot stop at asking whether a target sought help. It must also ask what evidence platforms preserve, how employers protect staff, how police respond and whether law can address harm that crosses platforms and borders or is amplified by AI.

“Tipping point” is not a boundary that can be precisely marked by one percentage. It is a warning that when online attacks repeatedly force women to self-censor, face offline risks or leave public roles, the harm extends beyond individual well-being to society’s ability to hear a range of voices. The report’s value and its sample boundaries should be kept in view together: it offers comparative evidence about public-facing workers in 2025, but does not speak for every country or every woman. Read it alongside The Chilling and MISA’s regional study in Southern Africa to ask how platforms, media organizations, governments and courts can turn support for women’s voices into transparent, appealable, multilingual systems that provide real remedies.