Abstract
Artificial intelligence (AI) is transforming adult content production by enabling users to generate highly customized sexual media through AI pornography platforms. Despite growing concerns about risks (e.g. illegal material generation and sex worker exploitation), AI pornography platform governance remains largely undocumented. We conceptualize AI pornography governance as fantasy arbitration, the process by which platforms determine which sexual fantasies can be technologically instantiated. Using an inductive content analysis of governance materials from 98 AI pornography platforms, we examine content prohibitions, moderation strategies and enforcement mechanisms, liability allocation, intellectual property rights, and privacy policies. Findings reveal heterogeneous governance regimes that create differential risk environments and illuminate how platforms independently dictate legitimate sexuality, determine who captures value from content generation, and commodify intimate data. Collectively, these governance practices position AI pornography platforms as corporate arbiters of sexual possibilities, with broader implications for intimate freedom, labor exploitation, and our co-evolution with technologies.
Introduction
Recent advances in generative artificial intelligence (AI) have given rise to AI-generated pornography, or AI porn. Through extensive customization options, AI porn platforms allow their users to create multimodal sexual media that aligns with their desires and preferences. Scholarship and media increasingly warn that these platforms may pose significant risks to individuals and society, including the generation of child sexual abuse material (CSAM), nonconsensual deepfake pornography, and violent content, as well as intellectual property violations and the exploitation of intimate user data. At the same time, AI porn offers potential benefits for research, therapy, education, sexual exploration, and sexual pleasure. Despite these implications, little is known about how AI porn platforms are governed, including the rules and mechanisms that structure their operations and shape their (un)ethical uses. Through an inductive content analysis of governance materials from 98 AI porn platforms, this study examines how governance decisions shape risks and benefits for users, performers, and society, while positioning platforms as regulators of erotic cultures and sexual behaviors.
To do so, this article first highlights the unique affordances of AI porn platforms and conceptualizes them as infrastructures enabling fantasy materialization: the technological manifestation of sexual desires. It then employs the Human-Erobot Interaction and Co-Evolution Model (HEICEM) as a framework to understand how selective pressures nested across ecological systems shape the co-evolution between erotic technologies and human sexuality, filtering which AI porn platforms, features, and sexual possibilities emerge and proliferate. Building on this framework, the article describes the role of platform governance, framing it as fantasy arbitration: the process by which platforms determine which sexual fantasies merit technological instantiation. It subsequently outlines the documented risks (e.g. CSAM, nonconsensual deepfakes, exploitation of sex workers) and potential benefits (e.g. sexual exploration, therapeutic applications, economic opportunities) associated with AI porn, establishing the stakes of governance decisions. This article then presents its methods and findings across five governance dimensions: (1) content prohibitions, (2) moderation strategies and enforcement mechanisms, (3) liability allocation, (4) intellectual property rights, and (5) privacy policies. Finally, it discusses how current governance practices shape the identified risks and benefits, revealing a heterogeneous landscape where platforms’ distinct rules, operational practices, and affordances differentially enable or prevent harms, while also exposing platforms’ broader power over intimate life and sexual possibility.
AI-generated pornography
Progress in generative AI has led to the proliferation of adult content platforms, such as websites and mobile applications like Candy.ai, Seduced.ai, or Pornplayground.ai, that enable the generation of AI porn. AI porn is typically defined as adult content generated or modified using AI tools (Döring et al., 2024, 2025; Lapointe et al., 2025; Yarmel and Lang, 2024). A recent content analysis of 36 AI porn platforms indicates that users can create and iteratively customize uncensored, realistic yet fictional images, short videos, and audio content (Lapointe et al., 2025). Some platforms also allow users to generate anthropomorphic erotic AI companions that can interact via text, audio messages, phone calls, and virtual reality (e.g. for sexting, role-play, or companionship).
Scholarship suggests that, compared to traditional pornography, AI porn offers unprecedented control and customizability over content creation, which reshapes how adult content is imagined, produced, and consumed (Döring et al., 2025; Lapointe et al., 2025; Yarmel and Lang, 2024). Rather than browsing pre-curated options, amateur and professional creators can now generate and refine content that reflects and evolves with their desires and preferences (e.g. setting, clothing, positions, behaviors, and physical appearances), from multimodal, photorealistic, and humanlike content to non-existent creatures, faces, and spaces (Lapointe et al., 2025). No longer passive viewers, users can thus act as both producers and performers within an AI porn ecosystem that allows easy access to increasingly tailored, user-driven, and interactive sexual media (Lapointe et al., 2025; Yarmel and Lang, 2024). Notably, whereas traditional porn production is limited by time, labor, equipment, and financial resources (Berg, 2021), these tools currently allow nearly instant mass production of erotic material at low to no cost.
These features position AI porn not as mere content distribution, but as an infrastructure enabling materialized fantasy: the technological manifestation of sexual desires and preferences (Lapointe et al., 2025). However, what gets materialized, how, and for whom is not random. Rather, we argue that it emerges from our technoerotic co-evolution.
Theoretical framework
The Human-Erobot Interaction and Co-Evolution Model (HEICEM; Dubé and Anctil, 2021) proposes that erotic technologies co-evolve with human sexuality through feedback loops: our desires drive technological development, which in turn reshapes how we access and experience sexuality. Applied to AI porn, this model suggests that platforms enabling materialized fantasies face multiple interconnected selective pressures that determine which features emerge, what content is permitted, and which platforms survive, which are nested across ecological systems. These include, for instance, users’ sexual desires and preferences, the infrastructures available to support these platforms (e.g. servers and payment processors), and the market competition with other platforms (microsystem). Other selective pressures may include the interconnected relationships that one’s partners or family have with pornography (mesosystem), the regulatory bodies that determine legal content and practices (exosystem), and the sociosexual climate of the countries they and their users operate in (macrosystem), which all change over time (chronosystems). Ultimately, technologies and technological traits best adapted to these pressures will endure, be replicated, and increasingly inhabit our (digital) world, while the others will be filtered out (Dubé and Anctil, 2021).
As such, the selective pressures exerted on AI porn platforms filter the content and systems we can access, the risks and benefits associated with these platforms, as well as the forms of sexuality we are able, or permitted, to have. Yet, little research has examined these pressures and their influence on AI porn. Thus, as an initial foray, this study focuses on one that intersects multiple systems: platform governance.
Platform governance
Platform governance refers to the norms, policies, and technical mechanisms that structure the management, functioning, and evolution of digital platform ecosystems (Duguay et al., 2020; Gorwa, 2019; Lessig, 2006). Formalized through documents, such as terms of service and privacy policies, governance defines permissible behavior, outlines the platform’s control infrastructure, and functions as contractual agreements between the service provider and its users (Gillespie, 2018; Plantin et al., 2018). Platform governance thus shapes normative expectations regarding rules, conduct, user safety, and user behavior, thereby playing a central role in determining platforms’ (mis)use and their emerging risks and benefits (Duguay et al., 2020; Gorwa, 2019; Lessig, 2006).
Given AI porn’s role in fantasy materialization, we posit that AI porn governance acts as fantasy arbitration: the platform’s determination of which sexual desires merit technological instantiation. Unlike content moderation of non-AI sexual media, AI porn platforms arbitrate both creation and distribution. This includes human-based decisions about and the moderation of prohibited content, which are enhanced or deployed through algorithmic processes. Namely, automated text filters moderate prompts by blocking prohibited keywords and concepts; interface design determines which options are available to users for generating content; and algorithmic systems determine what content can be generated through model training. By prohibiting certain materializations (e.g. nonconsensual deepfakes, CSAM) while permitting others, platforms adjudicate the boundaries of permissible fantasy even before they are materialized. In turn, such arbitration may influence sexual representations and scripts by determining which fantasies circulate widely, which remain marginalized, and ultimately what forms of sexuality inhabit our (digital) environments.
Shaped and enforced by socioculturally embedded legislation, payment processors, and technological infrastructures, governance practices thus position AI porn platforms as regulators of erotic cultures and individual sexual behaviors, determining which fantasies circulate and which are silenced. This underscores the broader stakes of AI porn platform governance: decisions about prohibition, moderation, liability, and privacy are decisions about the boundaries of sexual fantasy and cultural norms, becoming battlegrounds for competing politics of sexuality and what counts as (un)acceptable behaviors. These consequential decisions influence the potential risks and benefits of AI porn.
The risks of AI porn
The emergence of AI porn platforms has prompted growing concern among scholars and media outlets. First, many fear that AI porn platforms may facilitate the creation and access to illegal materials, such as depictions of CSAM and violent acts (e.g. murder or extreme gore; Shaughnessy, 2024; Nyce, 2024). In that regard, the Watch Foundation Report reported a 400% increase from 2024 to 2025 in actionable AI-generated content of child sexual abuse in the United Kingdom (Internet Watch Foundation, 2025).
Nonconsensual pornographic deepfakes have also become a focal concern (Döring et al., 2024; Ma’arif et al., 2025). Deepfakes are synthetic media (e.g. images, videos, or audio) generated or manipulated using deep-learning techniques to simulate a person’s likeness, voice, or actions. This technology can be used for entertainment or filmmaking (Karnouskos, 2020; Okolie, 2023). Yet, it mostly gained notoriety for its use in the creation of nonconsensual sexual content along with its consequences, primarily on women and girls (Chapman, 2024; Laffier and Rehman, 2023; Okolie, 2023). Its malicious use meant to humiliate, manipulate, or defame has led deepfakes to be classified as a new form of image-based sexual abuse and gender-based violence (Laffier and Rehman, 2023; Okolie, 2023). A central challenge lies in the hyperrealism of deepfakes and their rapid, uncontrolled spread. Once online, they can be repeatedly shared and are nearly impossible to eliminate (Chesney and Citron, 2019; Westerlund, 2019), targeting celebrities and private individuals alike and often causing severe psychological distress (Chapman, 2024; Laffier and Rehman, 2023).
AI porn platforms also pose risks for intellectual property and personality rights. Namely, the likenesses or original content of sex workers and private individuals may be repurposed into derivative creations without their knowledge or consent (e.g. deepfakes or AI characters modeled on real individuals; Butler et al., 2025). In addition, the bodies of sex workers may serve as the physical basis for deepfakes superimposing other people’s faces, reducing sex workers’ bodies to exploitable components for synthetic content generation (Macdonald, 2025; Pejcha, 2023). In parallel, the technical challenges of tracing such misuse likely leave performers without effective legal remedies (Anciaux and Gramaccia, 2025; Easterbrook-Smith, 2025; Pejcha, 2023). Moreover, output generated using likenesses and original content may be licensed and commercialized by platforms, potentially amplifying its nonconsensual dissemination. Meanwhile, the individuals represented (including sex workers who may be disproportionately targeted) are no longer legally recognized as the authors of the content, which undermines both their economic interests and their fundamental right to control representations of their own bodies and identities (Bendel, 2025; Butler et al., 2025; Henry and Witt, 2021; Macdonald, 2025; Pejcha, 2023).
Finally, AI porn platforms present risks related to data privacy. With the mass automation of personal data harvesting via AI systems, the risks of security breaches and unauthorized third-party access or sale have escalated substantially (Hu and Min, 2023; Murdoch, 2021). For instance, the Mozilla Foundation (Caltrider et al., 2024) found that 90% of 11 romantic chatbot platforms could share or sell user data, and that only half allowed users to delete their personal data. In addition, some contend that data may be collected without users’ full awareness due to vague or incomplete privacy policies, which prevent them from providing truly informed consent (Hu and Min, 2023; Usercentrics, 2025). These concerns are especially pronounced in the context of AI porn, where users may share highly sensitive content through exchanges with erotic chatbots (e.g. health and sexual preference data) that platforms can store, analyze, or repurpose (Caltrider et al., 2024; Deckker and Sumanasekara, 2025; Ragab et al., 2024). This sensitive content may be weaponized to coerce or manipulate people (Dubé and Anctil, 2021). AI models may also be trained on user data, increasing the likelihood of unintentional sensitive information leaks, even when the data are aggregated and anonymized (Hu and Min, 2023; Murdoch, 2021; Taeihagh, 2025).
Notably, as with other nonsexual platforms, these risks are compounded by capitalist incentives that push companies to rapidly deploy products and increasingly agential AI-powered systems whose objectives are misaligned with human well-being. This includes products and systems designed to maximize engagement and profits, despite the risk that they may limit content diversity, violate people’s consent, and lead to problematic behavior (e.g. crimes and suicide; Dubé and Anctil, 2021; Henry and Witt, 2021). Still, one way to partly address these risks is through governance. For example, governments and platforms can establish policies prohibiting deepfakes, violent content, and CSAM, along with mechanisms to report rule violations, remove content, and ban offenders. Yet, as with any regulations, challenges arise when it comes to enforcement and balancing protection with (sexual) freedom. Such freedom is important given that, despite its risks, AI porn also holds potential benefits.
The benefits of AI porn
AI porn may be leveraged to create and deploy user-tailored sex education and help individuals explore and learn about their sexual preferences (Dubé and Anctil, 2021). It may also be used to generate sexual content that promotes and represents sexual diversity (e.g. different bodies, demographics, and practices; Bergam et al., 2025; Döring et al., 2024; Lapointe et al., 2025) or varied sexual scripts that are less common in mainstream adult content.
Moreover, these platforms may be leveraged to produce personalized and controlled sexual stimuli for use in research and therapy, such as to test sexual responses to specific content or in progressive exposure protocols aimed at reducing sexual fears and avoidance patterns (Dubé and Anctil, 2021; Sanchez et al., 2024). In contrast with the risks previously described, AI systems may also facilitate safe(r), newer, and/or more lucrative sex work practices by allowing the creation and monetization of AI clones and (AI-modified) self-produced content (e.g. fictional persona on OnlyFans; Butler et al., 2025; Tangalakis-Lippert, 2024). These systems may also facilitate task automation and the consensual licensing of one’s likeness, offering commercial opportunities for passive income generation (Anciaux and Gramaccia, 2025; Easterbrook-Smith, 2025).
Importantly, these platforms may enhance pleasure by giving users greater control over the materialization of their sexual desires and preferences. But just as governance may partly mitigate some of its risks, it may also constrain the benefits of AI porn platforms (e.g. unintentionally prohibit consensual practices on moral grounds). Despite its implications, AI porn governance remains empirically undocumented.
The current study
Using inductive content analysis, this study examines governance materials from 98 AI pornography platforms. It then discusses the implications of this governance for human sexuality, fantasy materialization, as well as the risks and benefits of AI porn. By foregrounding governance practices, this research clarifies how AI porn platforms influence contemporary technoeroticism and increases transparency regarding their operations for users, researchers, and policymakers. As an initial step, these findings may inform evidence-based regulation and contribute to an AI porn ecosystem that prioritizes ethics and well-being in technological development.
Methods
Data gathering and sampling
Boolean search strings were used systematically across Google, Bing, and Yahoo to identify AI porn websites: “(AI OR Deepfake OR Deepnude OR Nudify OR Undress) AND (Porn OR Pornography OR Girlfriend).” These search terms were included because they represent the most commonly used marketing terminology and search queries associated with AI-generated sexual content, as identified through a scoping of the landscape. Websites were included if they enabled AI generation or customization of sexual content. Relevant websites were screened by two independent co-authors (VL and TK) on September 12th and 15th, 2024. The screening was limited to the first 2500 results across the three search engines. The co-authors identified 105 and 114 websites. Both lists were combined and reassessed, resulting in the exclusion of 16 websites that (a) did not include sexually explicit content, (b) redirected to a website already included in the sample, and/or (c) were inaccessible at the time of extraction. The final sample included 98 websites. On September 18, 2024, the terms and conditions, privacy policies, and other platform governance materials of each website were extracted and organized into individual Word documents to prepare for the analysis.
Sample characteristics
Seven (7.1%) of the websites did not provide any terms, policies, or guidelines to their users. Most platforms supported the creation of original image-based content (67.3%), 23.5% offered original video generation tools, 40.8% enabled users to modify existing content (e.g. deepnude, deepfake, nudifier), and 40.8% enabled the generation of erotic AI companions. By September 18, 2025, only 66.3% of the sampled websites were still active, while 6.1% had changed domain names, and 27.6% no longer existed.
Data analysis
To assess platform governance, the website data were coded and analyzed in accordance with the steps of inductive content analysis (Elo and Kyngäs, 2008). Two co-authors (VL and SR) conducted independent open coding on two websites, compared emerging categories, and reached an initial consensus. After independently coding the remaining websites, they discussed similarities, differences, and overlaps to develop a preliminary codebook, reviewed by two other co-authors (DL and SD). Categories were then organized hierarchically and defined. To assess inter-rater reliability, the coding scheme was applied to a randomly selected subset of AI porn websites (n = 10) by two coders (TK and VG), who assessed the presence (1) or absence (0) of each code. This yielded a Kappa of .76, below the targeted .80 threshold (McHugh, 2012). After discussing coding discrepancies, codes’ definitions and labeling were modified, resulting in a final number of 106 codes. Re-testing on 10 additional websites yielded a Kappa of .86. TK then coded the remaining 78 websites. The presence of codes was aggregated for each website to assess frequencies.
Results
Five first-order categories emerged, each pertaining to a different aspect of governance: (1) prohibited content and behaviors (i.e. what is permitted versus forbidden), (2) moderation strategies and enforcement mechanisms (i.e. how content is moderated and the consequences of policy or rule violations), (3) liability (i.e. who bears responsibility for the generated content), (4) intellectual property rights over content (i.e. who owns the content and who may profit from it), and (5) privacy policies (i.e. what data are collected, for what purposes, and with whom they may be shared).
Prohibited content and behaviors
Most websites (89.8%) outlined prohibited or permitted content and behaviors (see Table 1). A majority specified prohibited uses of the platform (75.5%), while others broadly prohibited any type of illegal content and behaviors (i.e. “no illegal content”; 14.3%). Specifically, 72.4% prevented the generation of materials that depict underage individuals, such as CSAM (64.3%; e.g. “lolicon,” “shota”) and the exploitation of minors (37.8%). Just over half of the websites (55.1%) explicitly prohibited the depiction of real individuals. A subset (35.7%) specified that content depicting real people could only be generated with formal authorization. Nearly half of the websites restricted depictions of physical violence (52.0%) and discriminatory or hateful content (50.0%). Nine platforms prohibited depictions involving bodily waste (e.g. feces, urine, vomit), while seven banned content related to cannibalism. Some also prohibited incest-themed content (8.2%), necrophilia (4.1%), or strongly ideological material, such as political or religious expressions (11.2%). A quarter of the platforms (24.5%) restricted depictions of sexual assault, including scenarios involving intoxicated, unconscious, or otherwise non-consenting individuals.
Prohibited content and behaviors.
For 7.1% of the platforms, no governance information was provided.
Additional restrictions included depictions involving weapons (15.3%), illicit drugs or pharmaceuticals (19.4%), non-human creatures (8.2%; e.g. aliens, vampires), fetish content (6.1%), bestiality (21.4%), and (para)suicidal behavior (22.4%). Counterintuitively, 22.4% prohibited the generation of sexually explicit material. In addition, several websites prohibited sharing or downloading content, allowing only in-site viewing (24.5%). One platform provided a comprehensive list of prohibited terms embedded in its automated filters (e.g. suffocation, drugs, daughter, farm sex, mutilation, castration).
Most platforms prohibited minors from using their services (85.7%), while 5.1% restricted access only to users under 13. Among these, four out of five required parental consent. In a few cases, platforms restricted access to other user groups, such as registered sex offenders (3.1%) and individuals (previously) involved in illegal activities (e.g. those charged with a crime; 3.1%).
Moderation strategies and enforcement mechanisms
Moderation strategies
Moderation strategies were commonly described as mechanisms to ensure lawful and appropriate use of the platform in accordance with community guidelines and applicable regulations (72.4%; see Table 2). Most platforms (67.3%) reported having a customer support team, with 5.1% offering 24/7 assistance. These strategies were either implemented directly by the platforms (49.0%), made available for users to apply themselves (65.3%), or both (42.9%). Moderation often included automated word filters designed to flag prohibited content (41.8%), human moderators manually reviewing content (35.7%), or both (29.4%). Moreover, 11.2% stated that content deemed unlawful could be manually altered by moderators to align with policy requirements. Two platforms reported using advanced age-identification technologies to detect CSAM in visual content (i.e. images or videos), and two others employed systems capable of blocking the generation of specific faces or names, such as those of public figures or individuals who had reported being targeted by nonconsensual deepfake pornography. These platforms invited victims to submit requests for their names or facial likenesses or to be excluded from outputs.
Moderation strategies and enforcement mechanisms.
Most platforms (65.3%) enabled participatory moderation. Specifically, 61.2% of platforms indicated that users could report content via a report or flag button (22.4%), direct messaging (52.0%), or a complaint form (28.6%). Moreover, two platforms required individuals who wish to create content based on a real person’s likeness to verify their consent by submitting a self-image that confirms their identity and authorization (addressed to performers or certified content creators). One platform published a transparency report detailing key content-moderation metrics, including the number of reported users, the total number of flagged items, and the amount of child sexual abuse material removed during the preceding quarter.
Enforcement mechanisms
Platforms generally outlined the consequences associated with generating prohibited content or non-compliance with their terms of use (90.8%; Table 2). Specifically, users were typically subject to permanent access termination (87.8%) or account suspension (67.3%). In addition, 61.2% indicated that they may report offending users to the relevant authorities. A few (6.1%) stated that users attempting to generate or disseminate illegal content may be subject to financial penalties. Some (39.8%) also reserved the right to remove any user content violating their content policies. Furthermore, 22.4% cited a repeat-infringement policy under which users who commit multiple violations would be permanently banned. Finally, 22.4% explicitly stated a zero-tolerance policy regarding illegal content (especially CSAM), specifying that any such attempt would result in immediate account termination and notification to law enforcement authorities.
Liability
Most addressed liability for both platform operation and available content (85.7%). A majority expressly (80.6%) stated that users bear full responsibility for the content they create using the platform, with 77.6% containing indemnification clauses, which required users to assume responsibility for any legal or financial claims arising from their use of the platform or the content produced. In addition, 46.9% of terms of service included waivers of legal rights. These clauses typically limited users’ ability to pursue legal action against the platform, such as waiving rights to participate in class actions or jury trials, or mandating arbitration as the only method of dispute resolution.
Intellectual property
Websites generally included statements regarding intellectual property rights and content ownership (83.7%). Of these, 65.3% recognized users as the sole owners of the content they generate, while 18.4% asserted ownership over the user-generated material. Most (83.7%) granted users a license to use their generated content. However, 40.8% restricted this license to personal, non-commercial use, and only 33.7% explicitly allowed users to exploit their content for commercial purposes. Conversely, 70.4% reported holding licensing rights over user-generated content, and 33.7% reserved the right to commercially exploit it, regardless of user ownership. Finally, 53.1% described a step-by-step procedure for submitting content-removal requests.
Privacy policies
Data collection and storage
Platforms often included a publicly accessible privacy policy (84.7%; Table 3). The majority of platforms specified the processing and retention of user-related information, including technical identifiers, such as device IDs and IP addresses (79.6%), while some stored financial data related to transactions (e.g. purchase history, financial institution; 30.6%). Some also processed inference data (30.6%), which allows the generation of user profiles based on behavioral patterns and preferences. Several platforms (40.8%) reported retaining user-generated content submitted via direct communications with platform representatives, while a few (12.2%) could access content from users’ social media profiles (e.g. posts, profile images, and in-app activity) when users authenticated via third-party services or shared website content externally. Moreover, 3.1% could access contact lists, either through linked social media accounts or directly from the user’s device.
Privacy policies.
Most websites (74.5%) processed basic identity data, including users’ names, email addresses, and dates of birth. A few (11.2%) indicated the storage of sensitive personal data, which included information pertaining to sexual orientation, political or religious affiliation, and ethnic background. A limited number (2.0%) stated storing health-related data, including sexual health status, medication use, and gender-affirming care. Approximately one-third retained user-generated content (31.6%), including uploaded images and chat transcripts. By contrast, 20.4% solely collected the information necessary for service operation. Over half of the platforms stated users’ rights over their data, such as the right to opt out, not to have their data transferred during corporate reorganization, or to request data correction (58.2%). However, only 44.9% of platforms explicitly acknowledged users’ right to erase their personal data permanently.
Processing purpose
Most websites (83.7%) retained user information to support operational functionality, for service optimization (e.g. analysis of usage trends, traffic patterns; 78.6%), to ensure legal compliance (74.5%), and/or for security-related purposes, such as fraud detection, account protection, and enforcement of platform integrity (69.4%). Only a few (3.1%) stated that user interactions may be utilized for model training purposes, 48.0% referenced data retention for corporate transactions (e.g. mergers, acquisitions, sale of business assets), and 54.1% collected and retained personal information for marketing and promotional purposes.
Data sharing
Most platforms (81.6%) typically disclosed the categories of third parties with whom user data may be shared. This generally included service providers acting on behalf of the platform (74.5%) and legal authorities in compliance with applicable law (74.5%). Data could also be shared with entities involved in corporate events such as mergers, acquisitions, or asset sales (46.9%), affiliated companies, and business partners (32.7%). None of the platforms explicitly stated that user data would be sold, but 38.8% affirmed that they do not, and will not, sell or share personal data for personal gain. The remaining 61.2% either omitted reference to data sales or provided ambiguous language insufficient to determine their practices. Finally, most platforms (73.5%) disclosed that they implement appropriate technical safeguards to protect users’ personal data, with 61.2% specifying that, regardless of the safeguards implemented, users should exercise caution when sharing data on the platform.
Discussion
Our intimate needs and pursuit of profit have fueled the emergence of a rapidly evolving AI porn ecosystem characterized by diverse features, rules, and safeguards. Findings reveal a structurally heterogeneous landscape characterized by divergent platform-level norms, operational practices, and technological affordances that, on paper, enable differential access to content and creative systems.
Content generation: prohibition, moderation, and enforcement
As frontline fantasy arbitration, prohibition and moderation shape which fantasies can be materialized, while enforcement mechanisms specify consequences for violations. Our analysis revealed that most platforms specifically prohibit CSAM, violent content, and nonconsensual depictions, implement moderation tools (e.g. user reporting, human review, automated filters), and state enforcement responses (e.g. content removal, account termination). This suggests that users are generally not granted unlimited freedom in content generation, and that safeguards are typically in place to prevent or minimize risks related to illegal or otherwise problematic material. However, three interdependent caveats remain.
First, this fantasy arbitration varies drastically across platforms. While most prohibit CSAM, violent content and nonconsensual depictions of individuals, other prohibitions diverge widely: some ban sexual violence depictions, fetish content and non-human creatures, and others provide minimal guidance (“no illegal content”) or none at all (Petit, 2025). Each platform thus operates as an independent arbiter with distinct rules and norms (DeCook et al., 2022; Pater et al., 2016). Definitional inconsistencies regarding forbidden content are further exemplified by the finding that several AI porn platforms counterintuitively prohibit sexually explicit content. This may be a deliberate ambiguity to maintain legal cover for hosting explicit material.
Second, reactive moderation allows fantasy materialization and circulation before intervention. Most platforms rely on community-based reporting, standard in the adult industry, but problematic for AI-generated content (Petit, 2025). Such strategies allow content to rapidly circulate and be downloaded before removal (Crawford and Gillespie, 2016; Gillespie, 2018, 2020; Henry and Witt, 2021), a concern amplified by AI porn’s unprecedented production speed and scale. Meanwhile, automated moderation faces high error rates, undermining reliability and risking removal of non-problematic content that constrains creative or otherwise ethical expressions of sexuality (Deckker and Sumanasekara, 2025; Gillespie, 2018; Gillett et al., 2022), while few platforms deploy other proactive safeguards like facial recognition blockers or age verification systems.
Third, technological affordances may enable violations regardless of stated prohibitions. Over a third of platforms’ core functionalities involve modifying existing content (e.g. nudifiers, deepfake generators). In the absence of proactive safeguards, these tools facilitate nonconsensual depictions of others despite platforms’ stated rules. Moreover, detecting whether generated content depicts real individuals without consent is technically challenging (Chesney and Citron, 2019; Westerlund, 2019), making reactive enforcement difficult and leaving such functionalities practically unconstrained. Notably, easy access to these functionalities lowers the technical and possibly the psychological barriers that might otherwise deter users from creating such content.
Overall, these limitations create governance heterogeneity with ecosystem-wide implications. Because AI porn governance depends on platform-specific norms rather than consistent industry standards, users can circumvent restrictive content policies by migrating to more permissive platforms (Ali et al., 2021; Rogers, 2020). Studies also suggest that being banned from a platform can escalate to more problematic behaviors on more permissive platforms (Ali et al., 2021; Innes and Innes, 2023). Individual platform governance may thus prove ineffective at minimizing the creation of illegal material without coordinated regulation, shifting the question from whether harmful content can be created to where it can materialize. Conversely, this governance heterogeneity may possibly mitigate restrictions on sexual freedom by allowing access to permissible but platform-restricted content on certain platforms.
Liability, intellectual property, and data concerns
Most platforms assign full responsibility for generated content to users and often state indemnification against legal claims (Rheins, 2023). In the context of AI porn, this strategy creates an asymmetry, where platforms provide and profit from an infrastructure enabling fantasy materialization (including tools that facilitate potentially harmful content), yet externalize liability for their misuse. The strategy’s effectiveness depends minimally on users’ understanding of what constitutes violations and their capacity to prevent them (Helberger et al., 2018; Suzor et al., 2017). Yet, many platforms undermine this foundation by providing vague or absent terms of service. In such cases, users cannot clearly identify prohibited content or effectively prevent violations for which they might be held accountable. Effective responsibilization requires transparent community guidelines that clearly communicate expectations and enable genuine user accountability (Gillespie, 2018; Helberger et al., 2018).
In addition, platforms vary widely in ownership structures. Some claim full ownership of uploaded content, others grant users ownership while retaining broad licensing rights to use, reproduce, and commercialize content. Users seeking to monetize their creations may find their work repurposed by platforms, while nonconsensual content may be inadvertently disseminated through platform licensing. Moreover, when users generate content based on someone’s likeness, users or platforms—not the individuals depicted—are recognized as content owners (Butler et al., 2025; Pejcha, 2023). This means that individuals’ faces and/or bodies can be appropriated for derivative content that users and platforms profit from without providing recognition or compensation (Butler et al., 2025; Macdonald, 2025). This is likely to impact sex workers disproportionately. In addition, only half of the platforms provide clear content removal procedures, limiting sex workers’ (or private individuals’) ability to address unauthorized likeness use. Such structures may extract labor power from performers while treating their bodies as exploitable raw material for platform profit in ways they neither control nor benefit from (Berg, 2021).
One platform provided a consent-based model in which performers license their likeness for user-generated content and can be compensated through the platform. This platform also allowed these individuals to keep their generations private for personal use. Such mechanisms may facilitate passive income and reduce labor demands by automating content creation and distribution, either through the site or other channels (e.g. OnlyFans, personal website; Anciaux and Gramaccia, 2025; Easterbrook-Smith, 2025). However, as Gorissen’s (2024) analysis of digital sex work under platform governance highlights the strong normative and structural control platforms exert over workers, it remains unclear whether AI porn platforms meaningfully protect performers’ rights, bargaining power, or long‑term economic security.
Finally, the privacy policies of AI porn platforms vary in data practices. While some collect minimal information necessary for basic functionality (e.g. login credentials, device IDs), a few gather far more extensive and sometimes sensitive data, including personal information (e.g. sexual orientation, religious beliefs), health-related details, and social media content (e.g. contact lists or private messages). These findings are consistent with prior research suggesting that romantic AI platforms collect sensitive data (Caltrider et al., 2024). Moreover, although no platforms explicitly stated that they would sell user data, over half included vague language regarding the possibility of data being shared with third parties. In such cases, user data may be treated as a commercial asset, reflecting the broader commodification of personal data, in which intimate information becomes a tradeable resource (Bottis and Bouchagiar, 2018). Combined with opaque governance and the immersive, interactive nature of these services (e.g. erotic chatbots), these data practices create fertile ground for user vulnerability and potential exploitation (Deckker and Sumanasekara, 2025; Taeihagh, 2025).
Implications for human technoeroticism
Through the lens of the HEICEM, our findings suggest that illegal content, such as CSAM, may be filtered out of most AI porn platforms—if rules are enforced, which remains unclear. The results also highlight that the current AI porn ecosystem is ripe for the proliferation of problematic content, such as nonconsensual deepfakes and violent depictions. Its limited moderation (e.g. word filters, human moderators, and user reporters) and easily circumvented consequences (e.g. access or account termination) are unlikely to prevent the spread of problematic content. By placing responsibility and ownership for content on users, platforms further shift legal risks that might otherwise incentivize them to regulate such content. Meanwhile, platforms can commodify intimate user data with limited transparency about what information is collected and how it is used.
In essence, our findings point to a central conclusion: the platform governance of AI porn likely enables many of its benefits but does little to mitigate its risks. Indeed, current platforms allow for tailored content creation, which may have beneficial applications in education, research, and therapy, as well as enhance access to sexual pleasure and companionship. They currently enable the materialization of most fantasies, via one platform or another. Yet, this landscape may also constrain sexual freedom to comply with morally infused yet ambiguous laws and regulations. These constraints may be amplified by payment processors’ enforcement of strict sexual content policies, which can constrain generative autonomy even in the production and consumption of ethical content (Webber and Franco, 2025). For example, about half of the platforms restrict depictions of physical violence, which can overlap with consensual sexual activities (e.g. Kink and BDSM). Moreover, little is done to curb risks related to data privacy, the nonconsensual use of performers’ content, and unethical or illegal content creation, including that which would be harmful to women, children, and sex workers. On the contrary, not unlike other (non)sexual platforms or services, the AI porn ecosystem is likely geared toward maximizing engagement and profits, even if it comes at the expense of human well-being.
Compounding this, platform governance exercises privatized regulation over our sexuality. As AI content achieves photorealistic fidelity yet remains synthetic, platforms must navigate an ontological ambiguity: can, or should, they prohibit socially condemned fantasies when they produce no direct victim? Given that platforms prohibit both illegal content and, in some cases, legal yet socially taboo content, our results suggest that platform governance may extend beyond victim protection to the normative regulation of desire itself, creating platform-specific regimes of sexual possibility (Gillett et al., 2022; Petit, 2025). This shifts governance from preventing tangible harm to policing imagination through anticipatory judgments about potential harms, societal effects, and moral acceptability (Petit, 2025). Therefore, these governance decisions are neither neutral nor purely technical but socially situated. For instance, while the moderation of CSAM is generally understood as essential, its expansion into a wide range of sexual representations raises concerns about a potential slippery slope, whereby harm-prevention logics become entangled with historically stigmatizing positions toward sex work and non-(hetero)normative sexualities. Although our study does not examine platforms’ ideological orientations directly, it remains unclear who governs these platforms, under what institutional constraints, and how anti-sex-work, heteronormative, or moralistic assumptions may become embedded in platform policy and enforcement.
Ultimately, through their role as arbiters of sexuality, platforms not only shape what content circulates but also whose bodies are exploitable, which fantasies are considered legitimate, who profits from sexual content generation, who bears the costs of violations, and how the digital artifacts of users’ fantasies are collected and (possibly) monetized. Rather than neutral infrastructure, AI porn platforms thus function as sites where commercial incentives and sexual regulation collectively exercise corporate authority over intimate life.
These governance dynamics extend to other nonsexual platforms but are made more evident by the sexual nature of AI porn platforms. Our findings about the governance of AI porn platforms—intersecting the platformization of sexual pleasure, the datification of our intimacy, and algorithmic moralities filtering (un)acceptable materialized fantasies—reveal larger power and incentive structures (Foucault, 1978). They show that, in an increasingly technology-mediated life, attempts at moral, ethical, or profit-driven control over our behavior and freedom extend into digital spaces. These attempts are especially visible on platforms designed to serve sexual purposes, thus offering a diagnostic site of larger tensions in the digital economy where political, corporate, and individual objectives compete and create the conditions of our co-evolution with technologies.
Limitations and strengths
This study has limits. For one, the presence or absence of governance information on websites does not mean that its practices are truly implemented (Henry and Witt, 2021; Ragab et al., 2024). Concretely, terms of service and privacy policies are performative instruments designed to satisfy legal requirements, manage liability, and shape public perception. Therefore, these texts may articulate ideal standards that exceed operational capacity and actual implementation.
In addition, qualitative analyses entail a degree of subjectivity in the development and application of the coding scheme. In some instances, categories were simplified or combined due to limited research team expertise or feasibility constraints, which may have reduced the specificity of the findings. The rapid evolution, high turnover, and increasing sophistication of AI porn platforms also create significant challenges in providing comprehensive, up-to-date results about this volatile ecosystem. Moreover, legal and regulatory frameworks, such as those pertaining to intellectual property and content prohibitions, vary widely across jurisdictions (e.g. the United States, the European Union, and China). These jurisdictional differences likely shape platform governance and may explain some of the heterogeneity we observe. That said, this study could not fully account for these differences as many platforms do not disclose their legal domicile in their governance information, and retrospective verification was not feasible given platform turnover (27.6% no longer exist). Our analysis thus describes governance as encountered by users through platform materials, rather than explaining jurisdictional determinants of governance variation.
In addition, the search terms targeted commonly marketed terminology but were not exhaustive. Platforms that use alternative descriptors or less conventional marketing language may have been excluded (e.g. creation tools for adult entertainers), potentially limiting the sample’s representativeness. Finally, the absence of comparative data from non-AI porn platforms limits our ability to determine whether the governance gaps identified are specific to AI-generated content or endemic to the broader adult entertainment industry.
Still, this study provides the first systematic examination of AI porn platform governance, addressing a rapidly expanding yet under-researched domain of sexology, media, and porn studies. It further highlights the complex tensions between innovation, commercial incentives, user safety, and the evolving place of AI porn platforms and their governance in our erotic lives and cultures. These insights also provide valuable implications for future research, policies, and regulations adapted to the age of generative AI and mass access to customizable pornography.
Future research
Future research should investigate how legal jurisdiction shapes AI porn governance practices and enforcement mechanisms. Research is also needed to understand user motivations for consuming AI-generated versus traditional pornographic content and to examine the types of content typically generated on these platforms, beyond stated prohibitions. Future research should investigate AI porn’s longitudinal effects on sexual practices, relational dynamics, and sexual scripts. Research should also examine how platform permissiveness shapes the content users generate and whether governance heterogeneity influences users’ platform choices. Importantly, as the ecosystem continues to expand and diversify, future research must develop effective safeguards against the misuse of AI-generated pornography while establishing robust standards for responsible deployment (Deckker and Sumanasekara, 2025). Achieving this requires interdisciplinary collaboration among researchers, policymakers, and educators to create adaptive and enforceable measures, including regulatory mechanisms, ethical guidelines, and public education strategies that foster informed and responsible engagement (Deckker and Sumanasekara, 2025; Kardos, 2025). Only through coordinated, context-aware governance standards can we establish the trust, accountability, and safety necessary for this rapidly evolving and highly consequential ecosystem. Understanding which governance mechanisms effectively prevent harm without constraining consensual expression remains essential to balancing protection and sexual autonomy in generative AI ecosystems.
Footnotes
Author contributions
Conceptualization: VL, SD, AP; Methodology: VL, SR, DL, SD; Data extraction: VL; Formal analysis: TK, VL, SR, VG; Writing—original draft preparation: VL, SD, AP; Writing—review and editing: All the authors; Supervision: DL, SD.
Consent
Not applicable.
Data Availability Statement
Declaration of conflicting interests
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Dr. Dubé consulted and acted as Scientific Director for the International Sexual Health & Wellness Research Institute–le Shaw (Byborg Enterprises). The authors have no other conflicts of interest to declare.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Ethics
Ethics approval was not required for this study, as it involves publicly available data.