1. Introduction
[1.1] The rise of accessible AI-powered image generation and synthesis technology in the 2020s has intensified the "Is it deepfake?" puzzle. Deepfakes—videos with AI-generated face-swapping and voice synthesization (Karnouskos 2020)—have made it increasingly difficult to distinguish genuine content from fabricated content. When beloved celebrities face photo or video scandals, fans often instinctively claim the content is "just a deepfake." While deepfakes have undermined the credibility of visual media, they are also highly believable, making them effective tools for disinformation. A few months before Taylor Swift's deepfake controversy in early 2024, a deepfake event occurred in the Thai Boys Love and Girls Love fandoms, where a celebrity video scandal opened Pandora's box around the creation, dissemination, and (re)presentation of deepfakes in this age of generative AI (GenAI). Girls Love (GL) is a genre of queer media featuring affective or physical intimacy between women, functioning similarly to Boys Love (BL) media, which focuses on male homoeroticism (Li and Pang 2025). We present acafans' methodological reflections on this event, which were not initially intended for publication. We begin by providing a brief overview of the context and our methodological approach. Then, we discuss fans' responses to the Thai GL celebrity video scandal, particularly how some created parodic deepfakes to undermine the authenticity of the scandalous video. This led us to an ethical dilemma: How should researchers engage with fans' creative practices that involve GenAI? Our reflections point toward a critically speculative ethics of care in fan studies amid a digital environment increasingly shaped by GenAI.
[1.2] Academic knowledge on AI has been black-boxed by research predominantly focusing on computer science, technicalities, and codes, leaving space for social sciences and the humanities to explore how machine learning reshapes our understanding of social relations and reality (Amoore et al. 2023; Browne et al. 2023). We contribute to that exploration by offering a methodological reflection on a deepfake event in fandom. In response to a scandal involving an alleged deepfake video in the Thai BL and GL fandoms, fans created a defensive parodic deepfake to discredit the scandalous video. While this fannish practice was well intentioned, it may have unintended and undesirable consequences. As researchers, we found ourselves caught in an ethical dilemma when seeking to (re)present the scandal and the parodic deepfake. Reflecting on these challenges, we argue that the "fans first" principle in fan studies needs expansion in a digital ecology where GenAI and machine learning are transforming fannish practices. Building upon the framework of fandom as human-community-machine interactions (HCMI) (Li and Pang 2024), which emphasizes the interplay between human and nonhuman agents within collective social and cultural practices, we propose a critically speculative ethics of care. This approach repositions humanity within an increasingly networked digital media ecology shaped by GenAI. However, it is important to note that while research on fannish parodic practices informs academic knowledge about fans' creativity, it may also inadvertently overlap with toxic practices employed by antifans—those who hate or dislike a celebrity—ultimately posing potential harm to both fans and the celebrities involved.
[1.3] As technology reshapes social relations, fan communities' responses to scandals highlight the complexity of how fans, as a community, scrutinize and interact with technologies and how those practices affect celebrities. This article may have begun as a set of methodological reflections; however, fans' navigation of the scandal raised significant questions about their perception of technologies. This presents an opportunity to rethink ethics in fandom research in an age where every click, every post, and every view are intertwined with AI and machine learning algorithms.
2. How it started: From a video scandal to a deepfake event
[2.1] One summer day in 2023, when we were doing our routine surfing and data collection for our research on the transnational fandom of Thai GL television series, a scandal involving two Thai celebrities broke out on the internet. A short video clip circulated on X (formerly Twitter) and Weibo (a China-based social media platform) showing a man and a woman sharing a brief kiss on the lips. Filmed from a distance, likely from outside through the windows into someone's home, the nine-second footage appeared blurry, vaguely showing both parties' faces (note 1). Speculation quickly arose that both parties in the video were Thai actors from BL and GL television series, queer media that fantasize about same-sex romance. The woman in the video was suspected to be Praew, a GL star who had a substantial international queer following resulting from the recent success of a GL series (note 2). The video sparked immense outcry and frustration within Praew's fandom, which caught our attention.
[2.2] The Thai GL and BL productions, known as Y-series, have become rising stars in the queer media industry, gaining phenomenal global popularity in the early 2020s (Baudinette 2023; Li and Pang 2025). Actors in these series, such as Praew, could amass millions of Instagram followers worldwide within just a few months. Having become two such followers of Thai GL and/or BL, we became acafans of Y-series, interested in its transnational fandom and analyzing social media data after obtaining ethics approval from our respective Institutional Review Boards (note 3). We have been observing posts and hashtags related to the Thai GL series, celebrities, and events, as well as fans' discussion in public posts on social media platforms such as X, Weibo, and Instagram. The discussion in this article is based on our reflections on the observation of social media interactions. Aware of the potential problems of lurking in digital fan spaces, we present our observations in aggregate form to ensure the anonymity of fans' discussions online (Markham 2012). In this capacity, we witnessed how this video scandal developed into a debate and play involving deepfake.
[2.3] We witnessed how this video upset many in the Thai GL fandom on social media. This reaction is understandable because GL media, akin to BL, featured not only same-sex romance in television drama series but also the fantasy of real-person ship (RPS) between actors who portray characters in romantic relationships on screen. That is, fans imagine (or expect) actors to be real-life, off-screen lovers. Likewise, since GL media often portrays a utopian world for queer women (Li and Pang 2025), fans are heavily invested in the queer fandom of the GL series in which Praew starred, as well as the RPS between Praew and her same-sex costar (note 4). Therefore, this video, although taken clandestinely and showing the two sharing a light, brief kiss on the lips, revealed shocking news to fans: Praew might have a male lover in real life. Such a seemingly small act caught on camera could potentially shatter fans' queer fantasy because the footage suggests that Praew is not a lesbian in real life (note 5).
[2.4] Many fans, most of whom are queer women, felt betrayed and dismayed that the RPS was not real; that is, the on-screen couple were not lovers in real life. As soon as the video circulated online, some fans quickly questioned whether it was real. This skepticism arose because spreading rumors about GL or BL stars engaging in heterosexual affairs has been a common tactic used by antifans to defame the stars and jeopardize their careers for being dishonest about their sexuality in real life. Among various speculations, the most notable was that the video was a deepfake, thus disqualifying it as credible evidence of Praew's sexuality. Such possibilities were seriously debated among fans until the official clarification statement was released three days later.
[2.5] Fans' frustration and denial of the authenticity of the video are not mere knee-jerk reactions to celebrity scandals; they manifest fans' deeper perception of technologies as well as their views of the Thai GL industry. In this particular scandal, fans' folk theories on social media reveal their perceptions and skepticism about how the media (or the machine) works, which inform their understanding of the incident (Ytre-Arne and Moe 2021; Xu et al. 2024).
3. Parodic deepfake as counter-information and digital play
[3.1] Within the three-day window between the video going viral and the release of the official statement from the artist management companies, we observed three folk theories that emerged on social media: (1) the video was real, but the person was a look-alike of Praew; (2) fans trusted that the management company would be able to prove the video was fake; and (3) the video was a malicious deepfake, in which somebody else's face had been swapped with Praew's. Among these, the third theory—that the video was a deepfake—circulated quickly. Deepfakes, created using AI or machine-learning tools that combine, alter, and overlay images and videos to produce realistic yet fake content, can be easily generated with open-source, user-friendly software and are designed for rapid dissemination on social media (Maras and Alexandrou 2019; Karnouskos 2020). Moreover, 2023 was a year where GenAI and deepfake were the talk of the town and became more accessible to ordinary people around the world. Researchers across disciplines have recently begun exploring the social implications of deepfakes, such as those related to legal contexts (Lau 2023; Maras and Alexandrou 2019), performance (Fletcher 2018), misinformation (Day 2019), and pornography (Popova 2020). Among these, large-scale, cross-national quantitative research by Ahmed (2023) revealed a telling result: deepfakes were more convincing than other forms of disinformation, and members of the public, despite being deceived, generally believed that they were less likely to be deceived by deepfakes than others. In the case of Praew, when the video could not be relied upon as evidence to verify or nullify information, further evidence was expected.
[3.2] Three days is a long time to wait for official clarification regarding the authenticity of the video; many fans could not wait. Some were eager to assess the realness of the video by conducting meticulous but relatively low-tech frame-by-frame dissection to look for traces of inconsistency and manipulation. Given the high technical threshold for deepfake detection, which involves training data for detection models (Groh et al. 2024; Jacobsen 2024), it is not surprising that we were not able to locate any conclusive video forensic report shared by fans. That being said, research on folk theories of false information suggests that people tend to believe information from their own social circles (Koçer et al. 2022). This could be why we kept encountering social media comments reiterating "the video is deepfake." In other words, fans reproduced and perpetuated the belief that the video was a deepfake without seeing any professional reports of deepfake detection.
[3.3] Since detecting deepfakes was more difficult than expected, some fans looked for alternatives to illustrate how easy it is to create them. Approximately one day after the outbreak of the scandal, we witnessed the circulation of a parodic deepfake made from the purported source video. Parody has been a prevalent, if not central, element of participatory media culture (Boxman-Shabtai 2018). Making and circulating parodic deepfakes in this case is a tactic that fans use to mock and undermine scandalous materials, thereby reducing the harm inflicted on both the fan community and the celebrity. The parodic deepfake we observed replaced the face of the woman, who was speculated to be Praew, with that of Shane, Praew's artist manager at the time. Shane was often seen at public events, so fans could easily find his footage and pictures on social media. The parodic deepfake showed a seamless face swap, closely resembling the facial expression in the original video. This parodic deepfake aimed to demonstrate that creating hyperrealistic deepfakes is not as difficult as people believe. We also noted that fans circulated and discussed this parodic deepfake primarily in a playful manner. This was partially because Shane was well liked by fans for his affability, which made the parodic deepfake relatable and allowed for queer pleasure in shipping him with the BL actor involved in the video. Additionally, the parodic deepfake served as comic relief for the fandom during a tough time following the scandal. Fans who saw the parodic deepfake jokingly expressed on social media that Shane and the BL actor "were a perfect match," and they would be happy to fantasize about them as lovers. With such a parodic video, the realness of the original video was also undermined.
[3.4] This form of AI-generated parodic image is unlike fans' other longstanding creative practices. Media users can quickly identify memes or GIFs that use celebrities' or public figures' faces for entertainment due to their recognizable style (Popova 2020). In contrast, the hyperrealistic parodic deepfakes made by fans risk disseminating believable disinformation, even if it was not the fans' intention to do so or if they did so out of goodwill. Here, the person whose face was swapped was Shane, an artist manager known to the fandom but not entirely a public figure or celebrity. Such playful yet believable and spreadable disinformation could have impacted him personally. Indeed, the parodic deepfake was not intended for circulation beyond the fan community. Still, it is difficult to ensure that they remain confined to specific internet spaces, given the ever-changing social media algorithms. Once these parodic deepfakes are circulated, they will be unerasable and retrievable online, subjecting all the individuals involved to doxing and cyberbullying. This is evident from what we have observed since the summer of 2023: the parodic deepfake images in this scandal were occasionally used by antifans to make fun of Shane and defame or attack Praew.
[3.5] More importantly, fans' tactics of countering difficult information or perceived deepfakes by creating parodic deepfakes might lead to unintended consequences for all parties involved or perceived to be involved. To be clear, we deplore paparazzi and stalkers and do not endorse the policing of celebrities' sexuality, but what sparked a debate between us about this incident is the ethics of representation in fandom research. Three days later, on their official X account, Praew's management posted a prerecorded video in which Praew apologized to the public and admitted that the video was taken without her consent. Without a doubt, any video taken without the consent of the parties involved should not be circulated or disseminated in any form—including academic research and publications. Nevertheless, how about the parodic deepfakes made by fans in which Shane's face was swapped? For pornographic deepfakes, where women (celebrities or ordinary people) are usually the victims, it is universally agreed that such images should not be disseminated by any means (Maddocks 2020; Story and Jenkins 2023). However, the parodic deepfake concerned here is more ambivalent. How should researchers study and represent the deepfaked body in academic research, and should they do so? Suppose we accept that digital images are subject to acceptable manipulation, such as applying filters, adjusting colors, or even removing objects with the help of AI-assisted functions in software; why should deepfakes be treated differently?
[3.6] We were caught in a dilemma. In research contexts, a digital image of the body includes both digital data and the material body, intersecting with complex ethical issues such as identifiability, confidentiality, anonymity, and publicness and privacy (Warfield et al. 2019). What distinguishes the deepfaked body from other digital images of the body is that the deepfaked body involves the bodies of multiple individuals, typically consisting of the body from the source video and the face from another body that has been swapped. To be more concise, the deepfake in our case includes Shane's face and Praew's body. This adds complexity in terms of both the ethics of engagement and the ethics of representation. In terms of consent for using fans' content, it would be challenging, if not impossible, to trace the source of the parodic deepfake and obtain permission from the creator and all the involved individuals. Deepfakes as digital artifacts complicate the conventional digital-material nexus of images collected from research participants, such as selfies and personal portraits. In terms of representation, how shall we analyze the deepfake images and disseminate them as research findings?
4. Rethinking the ethics and politics of care in fandom research
[4.1] When researching fans' engagement with deepfake, at least two levels of representations complicate each other. First, deepfake technology manipulates representations based on the already available visual and/or audio materials. Second, researchers writing about deepfakes and fans' responses in academic publications are also engaging in a form of representation. Taking this further, such scholarly representations potentially contribute to machine learning datasets. While we recognize the academic relevance and significance of discussing fans' perceptions and the use of AI in their responses, we are also acutely aware of the potential ethical issues related to representing the case.
[4.2] To ensure that fans will not be retraumatized as a result of our writing about the scandal, we turned to the literature on research ethics in fan studies. "Fans first," the foundation of ethical practices in fandom research, is a direct response to research that pathologizes fandom and disrespects fans' interests (Hellekson and Busse 2009). Good practices include not sharing direct URLs to fans' pages or fan fiction in publications and obtaining consent when using fans' work for research. Dym and Fiesler (2020) also encourage researchers to understand fandom's community and privacy norms. When using online fandom data, researchers should ensure proper consent from and acknowledgement of the fans concerned while obscuring identifiable personal information. Importantly, researchers should engage with fandom in a spirit of goodwill and reciprocity and ensure their research does no harm (Dym and Fiesler 2020). Similarly, The Association of Internet Researchers (2012) ethics working committee upholds the primary principle of "do no harm." Because fans have constantly been pushing legal and moral boundaries through their transformative works, it is particularly important to protect fan communities by not imposing harm and risks of harm through research, whether physical, psychological, financial, or through damage to reputation and intellectual property rights. Examples of harm in the context of fandom research include outing LGBTQ fans, exposing individuals to the legal ramifications of derivative work, and subjecting individuals or communities to online and offline harassment (Busse 2017). Nonetheless, harm in research should be defined contextually (Markham and Buchanan 2012). In the celebrity scandal discussed above, we should not only consider and respect fans' well-being but also that of Shane and Praew, even when Praew is a celebrity.
[4.3] AI-generated content can unexpectedly harm public figures and the public, whether intended as parody, satire, or disinformation. Glick (2023) highlighted incidents like the deepfake of Gabonese President Ali Bongo's 2019 New Year's video, which led to an unsuccessful coup, and deepfakes of Ukrainian President Volodymyr Zelensky used by Russians to demoralize Ukrainian soldiers. In entertainment, celebrities like Nicolas Cage and Bruce Willis have faced reputational harm from AI-generated content. Cage's image has been used in numerous memes, reinforcing his reputation as a "bad actor," while Willis's face was used in a Russian ad without consent, sparking political sensitivity during US sanctions on Russia (McGowan 2017). These cases demonstrate the undeniable harm caused by deepfakes.
[4.4] Without anyone being held accountable for the highly believable deepfake and its accompanying disinformation, celebrities' fear of losing control over their images and bodies is intensified, especially given the viral nature of social media. The long-term implications of deepfakes are unpredictable due to the lack of transparency in data usage for machine learning across various AI models (Cao and Yousefzadeh 2023). Unlike Busse's (2017) focus on text-based fandom harms such as fan fiction, Deller (2018) argues for considering a broader range of agents, including individuals, communities, media producers, public figures, and even physical objects. Given the networked nature of machine learning, we find Deller's approach more applicable to understanding harm in fandom research when AI is involved. In particular, in fandom, where the subject of fandom is an embodied person, harm could be inflicted not only on fans but also on public figures or celebrities.
[4.5] In the case of Praew discussed here, if we are to represent the parodic deepfake image in any visual form, it would be ideal—but almost impossible—to have obtained permission from the fan who created this deepfake image, as well as from all the individuals concerned, namely, Praew, Shane, and the BL actor involved. Even if we have the consent to do so for academic purposes, it is unnecessary to reproduce and display the images because a screenshot of the parodic deepfake would not add anything meaningful to our discussion—except perhaps a laugh at academic conferences—or serve as supplementary information in academic publications, since we can textually describe it along with other contextual information. It is unwise to provoke such a joke at the expense of potentially exacerbating the trauma experienced by the already affected fandom and contributing to machine learning datasets with disinformation. As explained above, online records of the leaked video and deepfake visuals have already been used for cyberbullying and doxing.
[4.6] This line of thinking aligns with the reflexive position of acafandom—crisscrossing academia and fandom—which is an implementation of fans first and do no harm in practice. It implies a certain level of care—the care for the fan community that a researcher identifies with or is part of. However, whether occupying such a position produces good scholarship and ethical practice has been much debated (Brooker et al. 2017; Hills 2012). This is because acafans have the power to not only represent fandom through their research but also silence the questionable aspects within fandom (Brooker et al. 2017; Busse and Hellekson 2012). Meanwhile, both fandom and social media are fragmented spaces, especially within the algorithmic culture of social media platforms, which often reinforce filter bubbles and echo chambers. The position of acafan does not guarantee understanding all the subgroups within a fandom. In the context of this scandal, our acafan position presented challenges. Emotionally, we were upset that the video was taken without Praew's consent, and we understood why some queer fans were distressed. However, we were deeply concerned when fans trolled and attacked Praew for "betraying" them, as fans jeopardizing actors' mental well-being is not uncommon in the Thai Y-series industry. This prompts us to consider the issue of care when fandom intersects with technology: What does the notion of care mean in the context of fandom research? As an overarching ethical framework, is fans first sufficient in the age of AI?
5. Toward a critically speculative ethics of care in fandom studies
[5.1] Given the highly commercialized and industrialized structure of the contemporary entertainment industry, celebrities are often treated as signs, texts, or signifiers. There is also a common assumption that celebrities are protected by capital and resources that make them invincible, physically and emotionally. We observe a contradictory development where celebrities increasingly emphasize their humanness, such as vulnerability, to appear more authentic and relatable. At the same time, the commodification of their bodies and personas simultaneously makes them prone to being perceived as objects of entertainment, subject to unreasonable demands in the name of professionalism, which dictates that actors should fulfill fans' fantasy 24/7, even in their private time and space. Similarly, in fandom research, we tend to celebrate fans' agency and participatory practices instead of addressing the experiences of celebrities and other individuals involved or affected. This focus is understandable, as fans are the primary subject matter in fandom research, and the entertainment industry often benefits from the fruits of fan labor, such as fame and popularity (Stanfill and Condis 2014). However, it is crucial to recognize that celebrities and other concerned individuals can be susceptible to various forms and degrees of harm, and that fandom consists not only of fans but other human subjects (celebrities and industry workers) and human-influenced technologies, such as digital platform algorithms and AI.
[5.2] After all, context matters. Willfully ignoring the significance of contexts in shaping perceptions of celebrity scandals and fans' responses risks replicating the coloniality present in academic knowledge production, which has been extensively critiqued in transnational feminist scholarship (Falcón 2016; Zerbe Enns et al. 2021). We asked ourselves: Is the video considered sensitive within its specific context, including the national/regional context and the context of a particular fandom? Compared to most Euro-American societies, where sexuality can be relatively openly discussed and celebrities can speak out for themselves, public displays of affection remain a taboo in many Asian societies. Even though Praew's leaked video was nonconsensual, the footage has entered the (semi)public domain on the internet. We often see similar scandals involving female celebrities' sexuality or intimate lives in Asia resulting in public apologies for disappointing their fans rather than defense of their own rights (note 6). This is why we debated how best to implement the speculative ethics of care in our writing while contributing to the emerging scholarly discussion on the broader impact of AI on fandom and society.
[5.3] We understand that researchers should commit to upholding academic integrity to provide sufficient contextual information for readers to understand the complexities of the issue and ground a critical discussion in our methodological reflections. This is particularly important because non-Western studies often necessitate more contextual details than their Western counterparts. Yet, this additional context creates a paradoxical challenge: it is difficult to include within the limited space of academic journals or presentations. Such details may appear overly descriptive compared to Western-focused work, despite their critical role in addressing the unfamiliarity of the non-Western setting.
[5.4] At the same time, we should minimize, if not eliminate, all possible risks of harm to the best of our ability. Indeed, readers may still be able to infer more details of the event from the academic footprint of the author; however, using pseudonyms and masking or removing direct identifiers are measures we should adopt as long as they do not conflict with the public interest. We argue that such ethics of care is particularly relevant in this age of AI, as digital creativity and artifacts made by both fans and antifans contribute to the existing and potential datasets in machine-learning models to train future AIs. The fan community and fandom research can contribute to such processes. Hence, participatory practices in fandom not only shape the semipublic fan communities but also influence social media algorithms and the machine learning dataset of AI.
[5.5] Building upon our observations on fans' entanglement with AI in response to Praew's scandal, we propose a speculative ethics of care for fan studies researchers investigating fans' interactions with AI-generated images. The word "speculative" here does not mean reckless claims but rather intellectual precaution that is imaginative, preemptive, and proactive. The speculative, as argued by feminist technoscience philosopher Maria Puig de la Bellacasa, indeed "connects to a feminist tradition for which this mode of thought about the possible is about provoking political and ethical imagination in the present" (2017, 7). Recent examples have shown that public figures can be victims of AI-generated visuals. Still, given the black-boxing of AI and machine learning, scholars in the humanities and social sciences can only critically speculate, with care, on the potential harms of technologies in relation to fan practices, such as parodic deepfakes and other AI-generated creativity (Amoore et al. 2023; Birrer and Just 2024).
[5.6] Care has been proposed by feminist philosophers as an alternative approach to moral philosophy (Keller and Kittay 2017). Care, in the broadest sense, refers to "a species of activity that includes everything we do to maintain, contain, and repair our 'world' so that we can live in it as well as possible" (Fisher and Tronto 1990, 40). The feminist ethics of care is, therefore, "a critical approach that seeks to understand the necessity of care to well-being, to understanding marginalization and identifying responsibility to remedy social injustices" (Brannelly and Barnes 2022, 6). While the feminist ethics of care have influenced a range of academic disciplines, such as healthcare, economics, and international relations (Keller and Kittay 2017), Puig de la Bellacasa (2017) contends that the definition of care is also a speculative project in its own right. The principle of fans first prioritizes the interest and well-being of fans, but when considering the intersectional and complex relationship between human subjects, community, and technologies in fandom, we have to think beyond fans first. Drawing upon Puig de la Bellacasa (2017), we articulate three guiding principles of the speculative ethics of care for fandom researchers that call for a rethinking of fandom research ethics beyond fans first:
[5.7] Thinking with fans: This refers to a relational way of thinking, which involves a commitment to a collective of knowledge-making (Puig de la Bellacasa 2017). This is closely related to fans first, as thinking with fans focuses on empathizing with them and staying attuned to the sociocultural trajectories of their desires and feelings while still addressing both the scandal and the potential harm it might bring to fans. This also means respecting fans' wishes to protect celebrities and others involved from further internet doxing and cyberbullying. One might be concerned that thinking with fans means foregoing academic rigor and integrity, but academic research is not merely a journalistic aggregation of fans' voices. Thinking with fans involves a thoughtful process of showing care to the community we research. In our case, we continually ask ourselves questions such as "How can we write about this scandal and fans' responses without triggering fans?"
[5.8] Thinking for fandom: Taking into consideration the feelings of fans, acafans can also contribute to fandom by offering alternative perspectives. Care often implies love and kindness, and so does the idealized perception of fandom as a loving, inclusive community that transcends linguistic and geographical barriers. However, fandom is also a site of conflict and disagreement. Considering an interdependent, relational web of relationships, fandom is increasingly filled with tension, especially on social media platforms where algorithms place us in filter bubbles and echo chambers. Researching fandom with care not only means thinking with and empathizing with fans but also being aware of tensions and dissent, hierarchies and power dynamics, as well as how existing relationality contributes to new forms of disconnection in relationships. Building care while recognizing divergent positions and offering new insights that may differ from them can be challenging, but we should not shy away from this responsibility.
[5.9] Thinking beyond fans and fandom: Fandom research should care for fans and beyond. One must respect everyone involved in the analyzed case, not only fans but also their subject(s)-of-fandom and other individuals, and see them as embodied human beings with feelings and dignity. Puig de la Bellacasa (2017) warns us of the danger of confusing ourselves as researchers with spokespersons who use marginalized others (such as fans) as arguments. In the case of leaked video and parodic deepfake we discussed above, the others include celebrities and the manager. As researchers, we must be aware of the multilayered representations at play, which complicate research of this kind. Studying fan engagement with deepfakes requires grappling with the represented nature of deepfakes themselves as well as the scholarly writing about them, which can feed into machine learning datasets and shape future knowledge. Hence, thinking beyond fans and fandom involves acknowledging what fandom means, as participatory culture has been transformed and reshaped by technology, as indicated by HCMI (Li and Pang 2024). In participating in fandom, we are also participating in the world of AI. Even though fans remain the core focus of inquiry in this field, researchers should shoulder the responsibility of speculating and minimizing the possible risks of harm to all concerned subjects.
[5.10] In addition to the basic ethical considerations and measures, such as obtaining approval from the institutional research ethics board and anonymizing the fans involved, we have taken a step further. As a praxis of the critically speculative ethics of care we proposed, we decided to pseudonymize the celebrities involved and not provide any screenshots or URLs to the images. It is critically speculative because no one knows precisely the extent and timescale of harm those images will cause in the age of machine learning and GenAI. Suppose both fans and celebrities consider the scandal traumatizing, there is no reason to exacerbate the situation by reproducing those images in the name of academic knowledge production while relinquishing ethics of care.
6. Conclusion
[6.1] Fans have always been the first among media users to experiment with new technologies. The current AI revolution is no exception. In parallel, fandom scholars also need to start addressing ethical issues when AI is involved in fans' parodic play and creative practices. From the puzzling question "Is it deepfake?" to the assertion "It is deepfake!" we have observed fans' approaching celebrity scandals by scrutinizing the authenticity of the scandalous video or using AI/face swap to counter such difficult information through playful disinformation. In other words, considering fandom as human-community-machine interactions (HCMI), when individual fans (human) and the broader fandom (community) are distressed by the video scandal, technology (machine) becomes the resort to verify or discredit the scandalous video. These practices, such as video dissection and the creation of parodic deepfakes, are further amplified on social media (machine) and engaged by fans (community). On the one hand, parodic deepfakes showcase participatory creativity. On the other hand, they pose potential risks, such as spreading disinformation and contributing to machine learning datasets, which could lead to unforeseeable consequences. In short, the interactions between fan communities and digital technologies play a significant role in the collective process of socio-technological meaning-making.
[6.2] Reflecting on the challenges of researching this deepfake event, we raise questions about representation and harm in fandom research within a digital ecology where AI and machine learning transform the very meaning of fandom as participatory culture. The proposed speculative ethics of care expands the fans first principle to include not only fans but also celebrities, industry workers, other human agents, and HCMI. With the emergence of deepfakes, fans' collective responses through digital play offer a significant area for exploration. Researchers must address the vulnerability of both fans and celebrities, the tensions within fandom, and the complex relationships among fans, celebrities, industry workers, and other social agents. We hope this article serves as a springboard for further discussions on ethics and care in fandom research concerning AI-generated playful practices, (dis)information, and the relevance of AI governance to fandom.
[6.3] Ethical dilemmas are not binary—to research or not to research, to write or not to write. In a digital landscape increasingly shaped by GenAI, data is both a source and vehicle of power (Hasselbalch 2021). Here, we have only begun to explore the ethical implications of researching fans' engagement with AI-generated content, and we acknowledge the constraints of conclusions drawn from reflections on a single case. While we are still waiting to see AI's more profound societal impact, our insights suggest avenues for future research on AI and fandom. For example, what if the parodic deepfake discussed is used for training future AI about Praew, GL, BL, or queer fandom? Recent developments, such as the introduction of AI-generated features on the wiki hosting service Fandom (David 2024) and AI-generated fanart and fan fiction (Lamerichs 2018; Mussies 2023), pose uneasy questions for fans: What if the products of fan labor, such as discussions and artwork, are used to train AI models that inform future applications? How might this affect fan communities, celebrities, and media producers? Further research is needed to examine how fans contribute to AI machine learning, the implications for fan communities, and the ethics of fandom research. Researchers should be aware of the potential harm caused by AI, deepfake, and online disinformation by conducting reflexive studies that uphold empathy and care.
[6.4] This deepfake event in the Thai GL fandom opens Pandora's box, prompting critical methodological reflections on research ethics despite the challenges it poses. The key to this critically speculative ethics is ensuring we do not harm fans and others who might be affected by our knowledge production. The three proposed principles serve as a starting point for reexamining fandom and participatory culture in the age of GenAI. When thinking with fans, researchers should envision themselves as the subjects in a deepfake scandal and critically reflect on whether they would be comfortable being treated and represented in that manner. In thinking for fandom, researchers, especially acafans, must consider the existing relationships that can lead to internal schisms, tensions, or disconnection, and navigate care within that context. Additionally, thinking beyond fans and fandom involves critically evaluating the potential impact of their findings when they are used to train AI models, which may affect fans and celebrities in ways that are not immediately evident. While this article primarily focuses on the Thai Y-series fandom, the ethical principles discussed here are not limited to specific fan communities. As fandoms evolve and become more complex, these principles can be extended, albeit with caution and critical awareness of the power dynamics at play. Understanding fandom in the HCMI framework (Li and Pang 2024) helps us better situate fannish practices within an evolving digital landscape. Meanwhile, it is crucial that we remain mindful of how our research might be used or misused, particularly in more problematic or harmful contexts. Thus, the responsibility to think ethically and reflexively extends beyond the confines of transformative fandoms—those that are generally supportive of the subjects they follow. This includes recognizing the ethical complexities in other types of fandoms, highlighting the importance of safeguarding both fan communities and broader social implications in a rapidly changing digital landscape. In this way, fandom research remains a reflexive project of coconstructing knowledge with fans—an endeavor that is not only intellectually rigorous but also ethical, healing, and empowering.
7. Acknowledgment
[7.1] We would like to thank Carolyn Pedwell and the anonymous reviewers for their insightful comments on an earlier version of this article. A previous version was presented at the Digital Feminisms in Asia Talk Series at The Chinese University of Hong Kong on March 12, 2025. We are grateful for Jia Tan’s invitation and the audience’s valuable comments.