IVE’s Yujin deepfake (Jan 2025)

In January 2025, Starship Entertainment, the agency behind K-pop group IVE, issued a public apology after its own official Weibo account reposted AI-generated deepfake content depicting member An Yujin. The incident stood out because the material was amplified not by an anonymous account but by the channel fans trust to represent the artist, briefly turning the agency into a distributor of the very abuse it polices.

What happened

On January 9, 2025, Starship's official account on Weibo, the Chinese social media platform, reposted content containing an AI-generated deepfake image of An Yujin paired with a derogatory caption. Fans noticed within hours, and criticism spread across Weibo and Korean fan communities. Starship initially explained that the employee managing the account had been trying to report malicious posts targeting the group when the content was shared, an explanation that did little to quell the backlash, according to The Korea Times.

On January 15, the agency issued a formal apology: "We sincerely apologize to An Yu-jin and her fans, who must have been deeply hurt by the severe negligence of an employee managing our Chinese social media account." Starship said the staff member received the company's highest level of disciplinary action and was removed from related duties, and it pledged to do everything possible to keep further harm from spreading, as reported by Soompi and The Korea Times.

Why this incident matters

K-pop idols are among the most heavily targeted victims of non-consensual deepfake imagery, and Korean agencies now run standing operations to monitor, report, and litigate against it. This case shows how fragile that workflow can be. Staff who handle reporting of synthetic abuse material interact with it constantly, and a single mis-click converts moderation into amplification. An official account also carries built-in reach and implied legitimacy, so a mistaken repost can expose millions of followers to content the agency exists to suppress. The episode reinforced calls for stricter internal controls around who handles abusive material, and for separating reporting workflows from the accounts that publish official content.

Where detection fits

Social media and fan-protection teams triage large volumes of suspect media, and screening files with an image detector before anyone engages with them reduces both exposure and mishandling. Much of this material is built with face swap techniques that leave statistical traces detection models are trained to catch, even when the result looks convincing to a human reviewer.