AI-fabricated witch hunt against Kim Soo-hyun may be 1st of many
Actor Kim Soo-hyun speaks at a press conference addressing controversy surrounding his private life at Stanford Hotel in Mapo District, Seoul, March 31, 2025. Korea Times file
Of the several false allegations made against actor Kim Soo-hyun, the most damaging came with the release of an audio recording purportedly featuring the voice of the late actor Kim Sae-ron, whose relationship with Kim Soo-hyun had become the focus of the controversy.
“We began dating when I was in middle school and broke up after I entered college. The first time we had sex was during winter break in my second year of middle school. Looking back now, I realize I was taken advantage of.”
The recording later turned out to have been generated using artificial intelligence (AI). It was among the false materials cited in the case against YouTuber Kim Se-eui, who was taken into custody on charges of spreading false information about Kim Soo-hyun.
In retrospect, things might have unfolded differently had the accusation been presented in another form. Had it been made in a written statement or delivered in person by Kim Se-eui, himself already a controversial figure, the public would likely have reacted with caution and demanded verification.
But it came in the form of an audio recording purportedly featuring the late actor’s voice, and the public immediately pounced. Suddenly, the difficult-to-believe accusation appeared to be backed by credible evidence.
The case demonstrated how false information generated using AI can be far more dangerous than the forms of misinformation that have rocked society in the past.
AI can sway opinion faster than police can verify
The false allegations against Kim Soo-hyun did not remain confined to sordid rumors circulating on YouTube. With the release of the AI-fabricated audio recording, Kim Se-eui’s accusation suddenly appeared to be an evidence-backed revelation, something mainstream reporters could justify covering and, to put it bluntly, something easy to write and quote.
An estimate by a civic group showed that 2,026 news articles were published over the eight days between March 10, 2025, when Kim Se-eui first made the accusation against Kim Soo-hyun, and March 17, 2025. In an attempt to contain the fallout, Kim Soo-hyun held a press conference of his own and denied the claims made in the recording. But few were convinced, largely because the voice sounded so real.
The danger posed by AI-manipulated false information did not end there. Even after an investigation began and authorities came to suspect that the material had been fabricated, proving it was another matter.
The National Forensic Service, in response to an inquiry from police, replied that it could not determine whether the audio recording had been manipulated using AI because the file submitted for analysis was not the original. Unless police can somehow establish how the recording was produced, proving that it was fabricated may be difficult, creating a gap that defendants can exploit.
Kim Se-eui, head of the YouTube channel HoverLab, arrives at the Seoul Central District Court in Seocho District, Seoul, May 26, for a hearing on an arrest warrant. Yonhap
For Kim Soo-hyun's case, police have concluded that the recording was fabricated based on the surrounding circumstances. While they could not determine with absolute certainty that the file had been manipulated, they reached the conclusion after considering its distribution trail, the existence of multiple conflicting versions, statements from those involved and signs that KakaoTalk messages had also been altered.
Ko Sang-rok, an attorney at Pil Law Firm representing Kim Soo-hyun, told the Hankook Ilbo that investigators likely took into account the fact that “completely different versions of a recording said to capture the same conversation on the same day and at the same time were distributed by the source to multiple people.”
For Kim Soo-hyun, luck was on his side. Others may not be so fortunate. The difficulties investigators face can ultimately carry over into the courtroom, where proving that evidence was fabricated may be even harder.
“In the past, forged or altered documents could be identified visually through handwriting, official seals and other features. But with evidence manipulated using AI, it may be difficult to determine how prosecutors should prove the fabrication at trial,” said Lee Chang-hyun, a former prosecutor and professor at Hankuk University of Foreign Studies Law School.
Lee added, however, that the rapid pace of advances in AI could also make such fabrications easier to detect. “With today’s sophisticated analytical technology, false or manipulated information should still be subject to sufficient verification and proof after the fact,” he said.
AI makes manipulation easier, experts urge faster safeguard
For now, the biggest concern is whether similar cases involving AI-manipulated evidence will emerge, a prospect that already appears to be on the horizon. In the past, altering someone’s voice required specialized equipment and considerable time. Now, a short voice sample and an online tool may be all that is needed to produce a convincing imitation.
That is why experts warn that the case should not be treated as merely another celebrity defamation case.
“The lid has been lifted on Pandora’s box, and not in a good way,” said Yoo Hyun-jae, a professor at Sogang University’s Graduate School of Communication.
“Kim Soo-hyun case just showed people with criminal intent how AI-generated material can be used to wage a battle for public opinion,” Yoo said. “There will undoubtedly be people who learned a new method of committing crimes from this case. Unless institutional safeguards are put in place quickly, we could see a second Kim Se-eui, and then a third.”
This article from the Hankook Ilbo, the sister publication of The Korea Times, is translated by a generative AI system and edited by The Korea Times.