May 22, 2025
SINGAPORE – TikTok’s Transparency and Accountability Center in Singapore on May 13, is a strangely quiet place for a platform that hosts the noise and energy of a billion users.
Here, the chaos of global short-form content is filtered, analyzed and — when necessary — stopped before it reaches a single screen. And, increasingly, among the digital debris being swept away are AI-generated deepfake videos targeting K-pop idols.
It is an issue whose urgency is increasing. In 2023, cybersecurity firm Security Hero published a report analyzing some 100,000 deepfake videos, the findings of which pointed squarely to South Korea.
Women, particularly female K-pop stars, were disproportionately targeted, with South Korean singers and actors accounting for nearly half of the explicit deepfakes examined. Of the top 10 most-targeted individuals, eight were Korean singers.
Agencies strike back — as does TikTok
The entertainment industry has since raised its guard.
Last August, major agencies like YG Entertainment and JYP Entertainment announced they would pursue all legal means to hold perpetrators accountable. In February, Hybe, home to BTS, took a further step, partnering with Gyeonggi Bukbu Provincial Police to combat the rise of cybercrimes, including deepfakes.
TikTok is under pressure too.
The app, owned by ByteDance, is not just a stage for dance challenges and fan edits. It is one of the most influential platforms for Generation Z across the globe — and therefore, a potential hotbed for misuse. To see how the company is confronting this responsibility, The Korea Herald visited the TAC facility in Singapore, where TikTok opened up about its back-end operations in its effort to demystify how content is moderated.
Moderation: AI first, humans second
The tour reveals a three-stage content review system that combines artificial intelligence with human oversight.
On average, about 1.6 million videos are removed from the platform each day, most of them by AI before anyone even sees them. The machine learning system does not just catch obvious red flags such as nudity or slurs — it is trained to detect context.
According to TikTok, one example would be how holding a cigarette might not trigger removal, but once the motion suggests the person is about to smoke, it is flagged as smoking-related content.
Similarly, a user holding a steak knife to eat would not trigger concern, but if the knife is held upright in a different context, it could be flagged as potentially harmful.
Local rules for local users
AI catches the vast share of problematic content, but some videos still slip through to the second stage: human moderators.
These reviewers, located around the world in the tens of thousands, assess videos based on TikTok’s community guidelines, taking into consideration local laws and cultural nuance. That sensitivity matters, especially in countries like South Korea, where users under 14 are banned from the platform and where parental controls like “Family Pairing” help families manage their children’s screen time and exposure.
In South Korea, TikTok enforces additional protections. Teen screen time is capped at 60 minutes by default. Parents can further customize viewing hours and restrict search keywords that appear in their child’s feed.
Evolving guidelines for evolving content
TikTok’s moderation rules are not static. The platform’s Trust and Safety team updates them in collaboration with regional experts, allowing moderation policies to adapt to emerging trends and cultural contexts.
At the TAC, a wall display outlines the nine sections of TikTok’s community guidelines. Six of them fall under the broader umbrella of “safety,” covering violent or criminal behavior, bullying, mental health, suicide and self-harm, adult content, and graphic violence.
The rest address misinformation, deepfake content, gambling, fraud and other threats. When it comes to generative AI — and specifically, deepfakes of public figures — TikTok enforces a zero-tolerance policy. That includes AI-generated content that impersonates celebrities in inappropriate ways.
Zero tolerance for AI exploitation
So how does TikTok specifically handle deepfakes of K-pop idols?
The company noted that any synthetic content violating its community guidelines, particularly involving impersonation or exploitation of public figures, is taken down immediately.
When it comes to sexual exploitation, TikTok enforces blanket removal regardless of whether the content is real or fake. In severe cases, where the situation warrants the involvement of law enforcement, TikTok will report the incident.
TikTok also emphasized that if content is found to involve sexual exploitation, it is deleted without delay.
Sexual content violations fall under two separate but closely related guideline sections: “safety and civic awareness” and “sensitive adult themes.”
TikTok explained that the distinction is necessary because, while some sexually suggestive material may fall under freedom of expression and is classified under sensitive adult themes, criminal behavior such as exploitation is treated under the safety and civic awareness policy with absolute seriousness and no leniency.
In the case of sexually exploitative deepfakes of K-pop idols, TikTok reiterated that such content is regarded as a severe violation of safety rules and is dealt with promptly and decisively.
jaaykim@heraldcorp.com