How to Stop AI Deepfakes: Strategies for Detecting, Regulating, and Preventing Synthetic Content
Artificial Intelligence & Machine Learning
In recent months, a new TikTok trend showcases users generating deep-fake videos of highly controversial historical figures—including Martin Luther King Jr., Adolf Hitler, and Charlie Kirk—saying completely falsified speech, justified by their humorous qualities yet underlined with racist ideologies. This form of content is only one example within the larger trend of “brain rot”. Short-form videos on social media, optimized for users to interact and engage, often lack meaning or substance; described by one 2024 lifestyle magazine as “brainrot humor”, the genre has “irreversibly warped the way people speak online”, ultimately “zooming from nothing to annoying super-saturation in an instant” (Press-Reynolds). In a similar fashion, the rapid development of artificial intelligence platforms like OpenAI’s Sora allows the democratization of creativity and artistic expression, with similar benefits in education; a recent analysis by two medical Ph.D. students found that artificial intelligence can “improve outcomes … from primary through higher and continuing education paradigms” (Monzon and Hays). Nevertheless, newer, emerging videos like the deepfake TikTok trend shines light toward the dangerous consequences of media generation platforms. With the ability to create anything imaginable, AI content could be used to spread misinformation, promote extremist ideologies, or even rewrite historical events. Thus, it becomes increasingly crucial to contemplate ethics: how can society effectively limit such ramifications of AI video generation without infringing on creative expression?
To determine how to treat AI-generated deepfakes, it’s first crucial to understand why they have become so widespread. Large corporations like TikTok and Instagram Reels prioritize user engagement over the information the content provides; their algorithms routinely push videos with both high and rapid like, comment, and share activity, regardless of purpose or content. Consequently, polarizing videos about sensitive topics are rewarded because of how audiences are pushed to interact, while possibly important but less shocking information receive less attention. Amnesty International’s work displays exactly that: through a manual research study involving simulated 13-year old TikTok feeds, the group found that “more than half the videos … related to mental health struggles” within 3 to 20 minutes on accounts signaling interest in mental health, with multiple videos “romanticizing … suicide” within an hour (Amnesty International). The feed’s display of videos “romanticizing suicide” indicates a complete ignorance of the value of the videos the algorithm suggests; rather, the lack of safety mechanisms toward what content is shown suggests a standalone focus on the amount of interaction. To apply this framework to deepfake videos, in one example, Martin Luther King is falsely illustrated inside a 7/11 holding a drink, in which he comments that his dream is that “one day, this [drink] will be free” (@aigettingreal) and sprints out of the store, as displayed in Figure 1. To some, this feels like trivial humor; to others, the deliberate choice of using a historical Black figure perpetuates negative stereotypes of African Americans, while simultaneously discrediting his major historical contributions. Extending Amnesty International’s analysis, videos like these representing extremist content are incredibly polarizing: between those justifying the content as “dark humor” and those angrily opposing MLK’s defamation, audience interaction is guaranteed.
Figure 1: A deepfake video of Martin Luther King holding a slushy, a popular 7/11 drink, and proclaiming his vision of making the drink free “for all who thirst” (@aigettingreal)
Thus, it is the innate quality of social media algorithms that drive this type of video to user feeds; to address the underlying issue, it’s imperative to regulate how platforms promote engagement-heavy videos.
Social media platforms also spread misinformation faster than authentic news, amplifying the extreme tendencies of existing algorithms. Such a phenomenon can be traced back to the emotional contagion effect, which argues that information backed by emotion spreads rapidly through social media platforms. The extent to which they propagate is also clear: a large-scale Twitter study spanning over 3 million people showed that false news was retweeted 70% more than true news, while made-up stories reached individuals 6 times faster than authentic ones (Naidu). Thus, since false information “evokes more surprise and disgust”, while true information “evokes sadness, anticipation, joy, and trust” (Naidu), it’s inevitable that outrageous, yet also fake, headlines and videos will attract more attention, and subsequently, be promoted by the algorithm. Drawing back to Amnesty International’s findings of the polarizing qualities of hyper-viral videos, the misinformation most picked up by the algorithm will likely be the most extreme variants, intensifying the strength and consequences of the spread of false news. Furthermore, artificial intelligence only makes this process easier. Its vast ability to fundamentally alter speech, video, and media is unprecedented in misinformation campaigns. Deepfake videos of MLK are both incredibly controversial yet completely falsified; consequently, they represent the strongest, most viral-like form of false information. Crucially, however, the “study determined that humans, not bots, were primarily responsible for the spread of false information on the platform” (Naidu); hence, rather than automated machines, it is the intrinsic human drive to promote such content that fuels the spread of false information. Although social media and AI regulation can diminish the effects of these videos, a complete rejection of AI-generated misinformation must first come from a change in how viewers engage and perceive such information—as it is more so the users, rather than the algorithms themselves, that push the content forward.
Policies targeting both social media and artificial intelligence have been promoted in large government settings. Nevertheless, in an effort to balance both free expression and the dangers of the above videos, the reach of the government is inherently limited. Specifically, in both Twitter v. Taamneh and Gonzalez v. Google LLC, the Supreme Court concluded that companies were not liable for the spread of terrorist content, as “their conduct did not amount to aiding and abetting an act of international terrorism” (“Social Media Algorithms: Content Recommendation, Moderation, and Congressional Considerations”). Although their ruling only
directly applies to terrorist-based content, the implicit consequences extend across all social media. If the spread of terrorist-based information—one of the most definitive misuses of social media—does not make the social media platforms liable, it naturally follows that companies have no responsibility to control racism, misinformation, and bigotry content. Consequently, lawmakers lack legal backing to regulate how social media algorithms promote negative information. Without a change directly stemming from the people, AI-generated deepfakes like these will continue to exert their influence within social media.
Beyond social media, another approach would be to regulate artificial intelligence as a whole, targeting the companies that provide the video generation service. Nonetheless, artificial intelligence policy research displays that the novelty of the technology makes AI regulation difficult. Fundamentally, Taeihagh, a Singaporean policy professor, remarks that societal challenges, international cooperation, power imbalances within companies, and the issue of free expression all contribute to the challenges in AI regulation. Specifically, “the general-purpose nature of generative AI and its applicability across various sectors and industries makes it difficult to define the scope of regulation and apply it consistently” (Taeihagh). AI video generation is a prime example of the “general-purpose nature of generative AI”. The field has implications across industries, including artistic expression, marketing efforts, education, and content creation, yet can also as easily be misused: the promotion of racist ideologies, as demonstrated by the MLK deepfakes, is a clear example. However, the wide applications of the technology make it extremely difficult to create regulation that doesn’t infringe on free will and expression; for instance, banning specific prompts reduces deepfake videos, yet also limits artistic potential. Thus, it’s incredibly difficult to delineate between the positive and negative use cases of artificial intelligence, and even harder to implement sufficiently specific policy that
doesn’t infringe on genuine expression. Although regulation can begin to limit the spread of AI-generated deepfakes, it’s clear that policy without a matching societal change is insufficient.
To determine the necessary societal change, it’s crucial to first understand why users even promote this form of AI-generated deepfake—or to that extent, any form of controversial, polarizing video. Thus, we look at the most primitive form of user interaction: the TikTok comment section. In the AI-generated video of MLK “robbing” a convenience store, comments range from uneasy pushback to high praise: nevertheless, the most liked are all positive in nature, supporting the video for its usage of dark, or black, humor. In particular, one user remarks that this was “one of the only times [where] ai was funny” (@chuckerioni), while others respond with photoshopped images of MLK, the laughing face emoji, and tagged user IDs. Only one comment rejects the racist remarks, suggesting that the video “is not funny”; instead, “it’s mad disrespectful and disturbing” (@iiam_zaizai). However, the next reply is an insult, hurling derogatory insults at the original commenter. Equally disturbing, the reply received almost double the likes than the original comment, indicating approval from other users toward the falsification of MLK for the context of humor. Through audience reaction, it’s clear that this idea of black humor is what propagates these videos through the algorithm. Innately, the very definition of the term is media that “juxtaposes morbid or ghastly elements with comical ones that underscore the senselessness or futility of life” (Britannica); however, the satirization and desensitization of a historically significant figure and situation becomes an indirect consequence. Artificial intelligence only makes this worse; with the ability to create any situation imaginable, what would traditionally be impossible to create can now exist, allowing the spread of increasingly controversial videos across the internet. As a result, by justifying these videos as dark humor, as seen through the overwhelmingly positive TikTok comments, users ignore
historical context and future ramifications; to limit audience interaction with AI deepfakes, we must first address the rationale behind “dark humor”.
Beyond the propagation of racist content, audience interaction with dark humor leads to an adoption of the ideologies that appear within the content. Namely, the flattening of serious situations appeals to cynicism: nothing is truly meaningful, and thus, everything, regardless of context, can be transformed into a joke. However, three Italian psychologists in a 2023 peer-reviewed study found that cynical humor is a “positive predictor of depression” (Dionigi), suggesting a link between dark humor and negative emotions. In a similar vein, three other European scientists in a recent study find that “lack of emotional clarity” and “impulse control difficulties were linked to greater depressive mood” when viewing depressive memes (Akil), concluding that internet “memes” centered around depression also influence the audience toward depression. Thus, both publications draw the same result: when dark humor is associated with negative emotions, users may also absorb the negative emotions into their life, both within their actions and their outlook toward the world. Given the effect of negative humor on audiences, it’s therefore likely that racist ideologies, even when presented in a satirical, humorous form like in the AI-generated deepfake videos display, can subconsciously result in the adoption of racist tendencies over time—a “bleeding” effect into the user’s life. Consequently, to combat such an arbitrary “flow” of racist ideologies, a rejection of the “dark humor” appeal is necessary.
However, dark humor is not intrinsically harmful. Kotlun, associated with Cal Poly, pushes back against Dionigi and Akil’s research in her absurdist analysis of “Rick and Morty”, where she suggests a difference between helpful and harmful dark humor. Viewed as a means to cope with the rapidly changing progression of the world, Koltun argues that the dark humor within “Rick and Morty” gives millennial viewers a “philosophical framework from which to
interpret the disorder that is postmodern America” (28), displaying a productive usage of dark humor: as a way of understanding and coming to terms with difficult, real-world situations. Nevertheless, the dark humor the AI-generated deepfakes fall short of such criteria. Rather than a defiance of the absurdness of the world, what differentiates these TikToks is their lack of valuable, insightful content. With no deeper meaning or inference, they serve as nothing more than an attack on history and a vehicle to promote racist ideologies, hidden under the cover of productive “dark humor”. It is exactly this distinguishing element of black humor that is crucial to acknowledge. The fight against the unproductive dark humor hence begins with a recognition of both the video’s value, historical context, and the “bleeding” effect: in an isolated space of rapid short-form content like TikTok, it’s easy to disregard the negative implications of seemingly purely humorous videos like these. Thus, change begins with continuous education and constant reminders of history, context, and critical thinking, both within the platform and in daily life.
It’s clear that policy cannot fix the underlying problem of human behavior; to ensure the negative consequences of video generation can truly be limited, education and awareness are key. Specifically, it’s imperative to ensure children, especially from a young age, are knowledgeable about dark humor, social media and the dangers of artificial intelligence. Not only will children grow into a generation where social media and technology will only rise in importance, they are most easily influenced by their surroundings; Vygotsky’s theory of childhood development describes how certain experiences and cultural norms impact adolescent development (McLeod). Likewise, Vygotsky draws on the concept of a More Knowledgeable Other (MKO): a person or tool that knows more than the child and teaches them core abilities that they use to generalize outwards (McLeod). Although this role is traditionally filled by
parents and teachers, the rise of technology and social media allows children to learn from what they experience on the internet; if children, subsequently, are immersed in deepfake content with racist messages, videos like these could become their MKO, possibly influencing what they grow to believe. As a result, as the internet culture shifts toward artificial intelligence and short-form content, it’s imperative for educators to both play a larger role and explore media literacy from a young age; without it, children know no better than to believe and engage with dangerous, AI-generated videos.
However, teachers and parents are not well-equipped to adequately address the rise in technology. Europe has taken action: the European Union’s Digital Education Action Plan remarks that fewer than “40% of educators across the EU felt ready to use digital technologies in teaching” (Digital Education Action Plan: policy background). Consequently, as technology inevitably improves, it’s crucial for the United States to adopt a similar policy plan focused specifically on artificial intelligence and the dangers of media generation, prioritizing critical thinking and promoting digital literacy. Likewise, media literacy education has been proven to be effective; a study of over 200 students comments that “the significant increase in knowledge … is in favor of the efficacy of such programs in improving students’ knowledge about media” (Geraee). In an age of technology where artificial intelligence is almost essential to daily life, the negative implications cannot be ignored; education is both effective and non-invasive toward creative expression, and can be utilized for both awareness of the negative consequences and recognition of the positive benefits. Thus, the necessary societal change begins with the younger generation; specifically, through AI-targeted education.
Nevertheless, education solely within schools is insufficient; continuous, recurring reminders are necessary for both children and adults on social media. Although algorithmic content is difficult to regulate, digital platforms can be urged to utilize AI video identifiers, warning labels for misinformation, and blurbs about historical context to ensure audiences fully understand the content they view. While scrolling through infinite feeds of social media content, it’s difficult to keep the dangers of video generation and misinformation at the front of every user’s mind; thus, constant nudging is necessary for a long-lasting societal change. Such labels are effective; even within the population of users who do not trust fact-checking applications, “warning effects are generally effective at reducing belief and sharing of false news” (Martel). Much of the worry in videos like the MLK deepfake is their promotion of both misinformation and racist ideologies; a recognition and clarification of both would give an unbiased and educated perspective for the users. Consequently, recurring messages on social media providing context and factual information are imperative to ensure education has an effect beyond the classroom.
Inevitably, artificial intelligence is here to stay. Although video creation has immense positive implications for a variety of sectors, such AI-generated deepfakes illuminate the devastating potential consequences of uncontrolled media generation. The real danger in artificial intelligence stretches far beyond trivial misinformation: as the technology continues to strengthen, complete histories are at risk of erosion because of the manipulation of photos, videos, and broader media. However, although regulation is necessary, it is ultimately insufficient. The widespread and decentralized nature of technology makes effective policy difficult, while, with no change, the emotional, polarizing nature of these videos will only continue to drive audience engagement. To truly address the dangers posed by AI deepfake videos, we must rather target the driving force behind the propagation of these controversial videos in the first place: namely, both social media algorithms and the misinformed culture of
unproductive dark humor. Rather than dismissing such videos under the shallow defense of “dark humor,” we should begin to treat them as a cultural medium, demanding attention to the historical context and value of information within each video. The first step is, then, to cultivate a cultural ethic of accountability and historical awareness through early education, both among adolescents and on social media applications. Indeed, to moderate the unrelenting growth of artificial intelligence, political action must be accompanied with a corresponding societal change.
Written by Jonathan Wang
Works Cited
Aigettingreal [@aigettingreal]. “He ran off with that slushie���� #mlk #ihaveadream #funny #sora #fyp.” TikTok, www.tiktok.com/@aigettingreal/video/7558921075824905485.
Akil, Atakan M., et al. “Exposure to Depression Memes on Social Media Increases Depressive Mood and It Is Moderated by Self-Regulation: Evidence from Self-Report and Resting EEG Assessments.” Frontiers in Psychology, vol. 13, June 2022,
doi:10.3389/fpsyg.2022.880065.
Amnesty International. “Global: TikTok’s ‘For You’ Feed Risks Pushing Children and Young People towards Harmful Mental Health Content.” Amnesty International, 7 Nov. 2023, www.amnesty.org/en/latest/news/2023/11/tiktok-risks-pushing-children-towards-harmful content/.
Britannica Editors. “black humour”. Encyclopedia Britannica, 28 Oct. 2025, https://www.britannica.com/topic/black-humor. Accessed 24 November 2025.
Chuckerioni [@chuckerioni]. Comment on “He ran off with that slushie #mlk #ihaveadream #funny #sora #fyp.” TikTok, uploaded by @aigettingreal,
https://www.tiktok.com/@aigettingreal/video/7558921075824905485
Dionigi, A., M. Duradoni, and L. Vagnoli. “Understanding the Association Between Humor and Emotional Distress: The Role of Light and Dark Humor in Predicting Depression, Anxiety, and Stress.” Europe’s Journal of Psychology, vol. 19, no. 4, Nov. 2023, p. Article e10013, doi:10.5964/ejop.10013.
European Commission. “Digital Education Action Plan: Policy Background.” European Education Area, last updated 18 Nov. 2025,
education.ec.europa.eu/focus-topics/digital-education/plan.
Geraee, Narjes et al. “Impact of media literacy education on knowledge and behavioral intention of adolescents in dealing with media messages according to Stages of Change.” Journal of advances in medical education & professionalism vol. 3,1 (2015): 9-14.
Iiam_zaizai [@iiam_zaizai]. Comment on “He ran off with that slushie #mlk #ihaveadream #funny #sora #fyp.” TikTok, uploaded by @aigettingreal,
https://www.tiktok.com/@aigettingreal/video/7558921075824905485
Koltun, Kim. “Rick, Morty, and Absurdism: The Millennial Allure of Dark Humor.” The Forum: Journal of History, vol. 10, no. 1, 2018, Article 12,
digitalcommons.calpoly.edu/forum/vol10/iss1/12.
Martel, Cameron, and David G. Rand. “Fact‑checker Warning Labels Are Effective Even for Those Who Distrust Fact‑checkers.” Nature Human Behaviour, vol. 8, 2024, pp. 1957–1967, https://doi.org/10.1038/s41562-024-01973-x
McLeod, Saul, PhD. “Vygotsky’s Theory of Cognitive Development.” Simply Psychology, 16 Oct. 2025, www.simplypsychology.org/vygotsky.html.
Monzon, Noahlana, and Franklin Alan Hays. “Leveraging Generative Artificial Intelligence to Improve Motivation and Retrieval in Higher Education Learners.” JMIR Medical Education, vol. 11, 2025, e59210, https://mededu.jmir.org/2025/1/e59210
Naidu, Priyanka. “Social Media and the Spread of Misinformation.” Museum of Sticks & Stones, 19 Mar. 2025,
www.sticksstones.org/exercises/social-media-and-the-spread-of-misinformation.
Press-Reynolds, Kieran. “How Brainrot Humour Infected the Internet with Surreal Gibberish.” Dazed, 25 June 2024,
“Social Media Algorithms: Content Recommendation, Moderation, and Congressional Considerations.” Congress.gov (CRS Product IF12462), 27 July 2023,
www.congress.gov/crs-product/IF12462.
Taeihagh, Araz. “Governance of Generative AI.” Policy and Society, vol. 44, no. 1, 2025, pp. 1–22. Oxford Academic, https://academic.oup.com/policyandsociety/article/44/1/1/7997395.
