Deepfake Technology and Gender-Based Violence: A Scoping Review

Terminology

The review indicated that deepfakes as a form of gendered violence have only recently been investigated in their own right, likely because they were not publicly available until 2017 and, more recently, due to the frequent integration of deepfake abuse with other forms of IBSA or TFSV. In reviews of TFSV, deepfakes are defined, and often included, as part of IBSA (Monteiro et al., 2024), featuring alongside non-consensual distribution of intimate images and sextortion (Henry et al., 2020). For example, deepfake gender-based violence has frequently been conflated with other types of IBSA such as photoshopped sexualized images (Huber, 2023a). There is a significant difference between photoshopped, and deepfake images. Deepfakes are technologically sophisticated products that look or sound more realistic than their photoshopped counterparts. Importantly, it is very difficult, and sometimes impossible, to prove deepfakes are indeed fake (Köbis et al., 2021; Korshunov & Marcel, 2020). This is not the case for photoshopped images, which can be shown to have been manipulated. This conflation of definitions was evident in Mumford et al.’s (2023) study examining the prevalence of victimization by TFSV in a young adult population in the United States (N = 2,676, aged 18–35), in which “digitally altered images” as a victimization item was operationalized as both photoshopped images and deepfakes.

There have been attempts to conceptually distinguish deepfake from other forms of IBSA. For example, Harper et al. (2021), in their paper on delineating non-consensual sexual image offending, distinguished deepfake media production from other forms of non-consensual sexual images, such as those known by the problematic term “revenge porn.” Similarly, Popova’s (2020) investigation of pornographic deepfakes of celebrities pointed to the differences between deepfake videos and other forms of digital blending of celebrity images. However, concomitantly, Paradiso et al.’s (2024) systematic review, which aimed to examine victimization and propensity to perpetrate IBSA, included deepfakes as a form of this behavior.

Overall, there remains a limited focus specifically on deepfake gender-based violence. There is, however, acknowledgement that it is an area requiring further research and one that could be a fruitful future direction for research into IBSA (e.g., Harper et al., 2023; Maddocks, 2018; Martínez-Bacaicoa et al., 2024; Plohl et al., 2024). We would argue that the falsified nature of deepfakes, along with the difficulties of evidencing their falseness, positions this form of IBSA differently to others. This is particularly the case in relation to the experiences of and recourses available to perpetrators, victim/survivors, and consumers as well as legislators and policy makers. This broad application of the term can also undermine the distinctiveness of deepfakes, which require complex machine learning algorithms to produce, and which pose unique ethical, legal, privacy, and security challenges compared to simpler forms of image editing. This misapplication of terminology may confound research efforts, as it fails to capture the specific risks and consequences associated with deepfake technology and AI. For these reasons, we would further propose that future projects should aim to design research that positions deepfakes, as opposed to fake content, as the object of focus, and explicitly explore deepfake gender-based violence.

Prevalence

Currently, research on prevalence as well as patterns that characterize deepfake victimization and perpetration is scant. Examinations of deepfake gender-based violence victimization are often investigated through a legal or regulatory lens, rather than from psychological or criminological ones (see, for instance, Umbach et al., 2024). This, as mentioned in the previous section, is further complicated by the conflation of deepfake gender-based violence with other forms of IBSA, particularly “revenge pornography,” when assessing prevalence statistics. Bansal et al. (2024) have argued that, given the existing conflations, it is difficult to establish the prevalence statistics of deepfakes specifically, while a scoping review of IBSA by Henry and Beard (2024) found no prevalence statistics for the perpetration of creating deepfake images, as these were embedded in the statistics for IBSA. This is further evidenced in Turner et al.’s (2025) research on prevalence and characteristics of perpetration and victimization in IBSA in the USA. While deepfakes were referenced as a form of image-based sexual exploitation alongside photoshopping, these different forms were included together under a more general focus on the perpetration of non-consensual image sharing. This pattern is visible in Sanders et al.’s (2023) U.K. mixed-method study on IBSA experiences of commercial creators whose content has been misused. In Sander et al.’s study, deepfakes were included within the definition used; however, due to conflation with other forms of IBSA, there was a lack of study of the specificity of deepfake victimization. While a recent review by Sheikh and Rogers (2024) on TFSV research claimed that 6% of women in South Asian countries experienced deepfake harassment, the statistic arose from a conflation of deepfakes with “synthetic porn,” which included photoshopped pornographic images. Deepfake gender-based violence was not specifically distinguished. Indeed, the relationship between impersonation and deepfake in the above study appears to be derived from an article from a media and entertainment company (Ruíz et al., 2020). Thus, the literature indicates that there is a propensity to mention or acknowledge deepfake as a form of image-related gender-based violence without a specific focus on it as a topic of study in its own right.

This lack of deepfake specificity also extended to cross-cultural research. Rackley et al.’s (2021) 3-year cross-jurisdictional project with quantitative surveys and qualitative interviews across the United Kingdom, New Zealand, and Australia assessed the prevalence, nature, and impact of IBSA. In this study, deepfake was mentioned alongside other forms of IBSA (e.g., fakeporn and Photoshop), however, excerpts from interviews with 75 women on their experiences of image manipulation, discuss Photoshop rather than the use of deepfake technology per se. While it is certainly possible that Photoshop and deepfakes are used interchangeably in everyday discourse, the lack of clarifications means that it is not yet possible to separate findings specifically about deepfake from other forms of IBSA.

Exceptions to the above pattern, which have specifically focused on deepfakes, include a recent doctoral thesis (Bahrenburg, 2024) located in psychology, which examined demographics of people victimized by different types of cyber sexual abuse (N = 126, 82.5% aged 18–26). This research found that victimization through deepfake pornography was the least common form of abuse (3 participants, 2.4%). Data collection materials for this study, however, defined deepfake as an “image superimposed on a sexually explicit video or image to appear as if you are acting out the sexually explicit material” (p. 82), so as including all forms of image manipulation. Therefore, it is likely that participants may have reported other forms of IBSA instead of, or alongside, deepfakes specifically because of the conflation of terms in everyday sense-making as argued above.

Taken together, this review suggests that there is a considerable gap in current knowledge on the prevalence of victimization–perpetration as well as methodological procedures that enable respondents to distinguish deepfake gender-based violence from other forms of IBSA.

Social Perceptions and Treatment of Deepfake: Sharing and Responsibility

Social media has been identified as playing a key role in public exposure to non-sexual deepfakes, and the often-comical nature of the deepfake content shared on these sites appears to redirect responsibility away from online platforms. Indeed, there is some suggestion that the media has contributed to the promotion of deepfake gender-based violence as less important than deepfakes used for disinformation (Gosse & Burkell, 2020). However, a study that examined online audience responses to witnessing “deepfake pornography” found that while audience engagement varied, anger was more strongly associated with bystander intervention than guilt (Wang & Kim, 2022). One survey found that awareness of deepfakes is determined by intensity and type of social media use, and that the level of humor of the deepfake content can negatively affect perceived responsibility of platforms to regulate deepfakes (Cochran & Napshin, 2021). There is a lack of confidence in technology being able to solve the problem of deepfakes (Cochran & Napshin, 2021), which attributes responsibility to individuals. Further cyberpsychology research on individual responsibility attributions for creation and sharing of non-sexual deepfakes found that people externalize responsibility for deepfake images. For example, younger people who find deepfakes humorous are more likely to perceive themselves as responsible for sharing this content (Napshin et al., 2024). Such studies may support further research into bystander response to audiencing or sharing deepfakes on social media platforms, including slowing the spread of distribution. This is an important area, with educational implications (Ali et al., 2021).

Alongside responsibility considerations around sharing deepfakes, experimental studies have examined social judgements of sexualized deepfake images. Fido et al. (2022) found that people have more lenient judgments of sexualized deepfakes if the image is of a celebrity, of a man, and if the image was for self-sexual gratification as opposed to being publicly shared. Fido et al.’s study points to the gendered nature of deepfake victimization, the sexual double standards for public figures, and the influence of social perceptions on public sharing of sexual deepfakes. In a similar vein, Maddocks (2020), found, in a sample of 3000 tweets about or on deepfake content, that there was a tendency for this form of gendered violence to be overlooked, marginalized, and automatically distributed by bot accounts. Taken together, these studies point to gendered inequalities not only in patterns in victimization–perpetration but in wider social and cultural responses to deepfake as a form of gendered violence. A recent study by Harper et al. (2023), which developed the Beliefs about Revenge Pornography Questionnaire, suggested that this tool could be developed into an adapted version to examine beliefs about deepfake gender-based violence. However, at the time of this study, there are no known measures of beliefs about deepfake offending.

Deepfake Perpetration—Creators and Consumers

Research examining motivation for the creation and consumption of deepfake images is in their infancy. While not explicitly on deepfakes as a form of violence, Marini et al. (2024) conducted a cognitive psychological experiment concerned with the authenticity of content in relation to consumption of “deepfake pornography.” In this experiment, 27 women and 30 men were asked to view real pictures of men and women models wearing lingerie or swimsuits and were told that the images were either of real people or artificially generated. The findings from this experiment suggested that perceived realness of the images was highly positively correlated with participant’s self-reported sexual arousal. While the rationale for this experiment was not explicitly related to questions around motivations of perpetration or consumption of non-consensual deepfake gender-based violence, it may be of relevance to this issue. This is because the findings may point to actual experiences related to, or broader social acceptabilities around, the relationship between deepfake viewing, sexual gratification, and talking about such consumption.

The purposes of deepfake creation have been recognized as a highly complex phenomenon (Tanwar & Mahapatra, 2024). Frameworks for understanding offline sex offending have been applied to online gendered violence, including deepfake creation. In common with Marini et al.’s (2024) conclusions, the idea of sexual gratification as a motivation for deepfake creation was highlighted in Harper et al.’s (2021) review of non-consensual deepfake creation as sex offending. Drawing on the sex offending literature, Harper et al. (2021) argue that motivations for deepfake creation can be linked to the underpinnings of other forms of sex offending, for example, dysfunctional sexual scripts, sexual compulsivity, and self-sexual satisfaction. More generally, Vasist and Krishnan (2023) argued that creators of sexualized deepfakes are often represented in research as malicious in their intent to, for example, enact reputational damage, control, and exact revenge against those depicted in deepfakes. However, Vasist and Krishnan (2023) note that a focus on motivation alone is not a sufficient explanation for deepfake creation. This is because deepfake creation is becoming a highly stigmatized and criminalized practice across various jurisdictions globally. As a result of these burgeoning legislations, open-source methods for creating deepfakes are becoming less accessible. However, the challenges of creating deepfakes have decreased, allowing creators to produce deepfake gender-based violence content with only limited technical expertise.

Moving away from individualized accounts of motivation, there has been a small number of studies that have explored deepfake creation and consumption in online communities. Popova (2020) explored communities that engaged with digitally created sexual content of celebrities, which included deepfake celebrity videos. Popova was particularly interested in the ways in which authenticity and intimacy become relevant to audience engagements that seek to access the “real” person outside of the celebrity image. Popova argues that while some sexualized content such as celebrity hacks are related to accessing something real or authentic in the lives of celebrities, within particular communities’ engagement with digitally altered sexualized images is located within broader practices of fictional imaginaries (akin to fan fiction). Deepfake sexualized content of celebrities is similarly not tied to the lives of the celebrities themselves, according to Popova. Instead, they are tagged as fake, unreal, and artifacts of digital alterations of both celebrity and porn and are intentionally contained within the community rather than distributed more widely, as in the case of celebrity hacks. In these cases, Popova argues, deepfakes appear to be for “relatively private pleasures shared only with those who have a similar understanding of the material” (p. 21).

Communities associated with deepfake creation were also the object of study by Newton and Stanfill (2020) who examined discussions of deepfakes on Github, which is a platform for software development and primarily androcentric. Discussions were located in a broader culture of toxic “geek” masculinity in which technological expertise is harnessed to perform hegemonic masculinity, while at the same time, claiming a marginalized masculine status. The creation and discussion of non-consensual deepfake pornography were accomplished by abstracting creation from the human, disavowal of pornographic elements, and an indifference to harm. However, the social unacceptability of deepfake creation practices was recognized by disguising creator identities. Newton and Stanfill (2020) conclude that these patterned ways of discussion suggest that “deepfakes serve as a site for using technology to enact hegemonic masculinity’s dominance while hiding behind technocratic rationality to refuse empathy” (p. 11). Similar patterns were observed by Winter and Salter (2020) on community interactions on deepfakes on Reddit and Github and are consistent with Kalpokas et al. (2024) review on non-consensual deepfake creation and sharing, which reflect an underlying culture of toxic masculinity. This is also resonant with Ayers’ (2021) work on how deepfake face swaps perpetuate white male hegemony and masculine power in YouTube videos, echoing the androcentric and patriarchal power imbalances located in deepfake creation. Deepfake gender-based violence has thus been conceptualized as a medium through which creators can perform a version of hypermasculine violence and is also concerned with entitlement enacted through the coming together of online communities for sharing non-consensual synthetic sexual imagery (Kalpokas et al., 2024). This entitlement to procure deepfake non-consensual imagery is flagged in Kikerpill (2020) who analyzed a thread on a community site that was dedicated to deepfake “designer pornography” of a person they had a relationship with during secondary school.

The research on online communities identified in our scoping review points to the ways in which creation and consumption are embedded within social practices. This makes particularly relevant Hancock and Bailenson’s (2021) call for future research focusing on social issues around deepfake creation, while keeping updated with evolutions in the capabilities of these technologies. For example, they suggest that this could include investigations of deepfakes used for live video as opposed to pre-recorded video, which is the form of deepfake that is predominantly discussed in current literature.