AlgorithmWatch Frames Sexualized Deepfakes as Gender-Based Digital Violence

The organization links non-consensual AI-generated sexual imagery to harassment, abuse, and coercion that can extend from public platforms into private chats and offline life.

AlgorithmWatch has published a guide defining sexualized deepfakes as one form of digital sexualized violence, rather than a standalone AI misuse problem. Its framing places the creation or distribution of non-consensual sexualized deepfakes within a broader pattern of gender-based abuse carried out through technology and digital services.

The organization says there is no single, uniform definition of digital sexualized violence. It adopts an approach used by the Federal Association of Rape Crisis Centres and Women’s Counselling Centres, known as bff, which treats digital violence as an umbrella term for gender-based violence committed with technological means or in digital spaces.

Under that approach, digital abuse is not separate from offline abuse. It can extend existing sexualized violence into social media, messaging services, closed groups, fake profiles, and other online channels. For compliance teams and platform operators, that distinction matters because harmful conduct may cross several products and contexts rather than appearing as one isolated post or image.

AlgorithmWatch includes sexualized deepfakes made or shared without consent among several forms of digital sexualized violence. The list also includes unsolicited sexual images, non-consensual sharing of intimate images, sextortion, sexualized insults and threats, cybergrooming, and stalking or doxxing with a sexualized component.

Image-based sexualized violence includes both the sending of unsolicited sexual imagery and the distribution of intimate material without permission. Sextortion involves blackmail using intimate content. Cybergrooming describes the targeted initiation of sexual contact with minors online. AlgorithmWatch also identifies sexualized stalking and doxxing as potentially escalating harms. Doxxing can involve publishing contact information or creating fake profiles, sometimes alongside calls for violence.

The organization emphasizes that sexualized deepfakes should be assessed through gender and power dynamics, not only through questions about synthetic media or impersonation. In research and academic discussion cited by AlgorithmWatch, non-consensual sexualized deepfakes are treated as gender-based violence embedded in patriarchal power relations.

Women and children are particularly affected, AlgorithmWatch says. It also identifies several groups as disproportionately targeted: women with migration histories, girls, children and adolescents, and queer and trans people. Women in public-facing roles, including actresses, singers, politicians, and activists, are frequent targets of non-consensual sexualized deepfakes because images of them are often widely available online.

That risk profile has practical implications for moderation and reporting systems. A platform policy focused narrowly on manipulated media may miss related conduct, including threats, repeat harassment, publication of personal information, non-consensual intimate-image sharing, and coercion in private communications. Policymakers and educators may likewise need prevention programs that address gender-based abuse and vulnerability, rather than treating deepfakes solely as a technical literacy issue.

AlgorithmWatch presents its guide as including self-defense information for people affected by digital sexualized violence. However, the available material does not establish the guide’s specific recommended steps, such as evidence preservation, reporting routes, account-security measures, or legal-support options. Readers should not infer that any particular response procedure is endorsed until the full guidance is available.

The central compliance takeaway is clearer: AlgorithmWatch’s definition sets a broader scope than deepfake detection alone. Systems for reporting, enforcement, victim support, and repeat-offender detection may need to account for abuse that spans public posts, direct messages, closed groups, impersonation, and the non-consensual circulation of sexualized material.