Correction (19 October 2026): An earlier version of this article described events that we could not verify. It has been rewritten from verifiable sources, which are linked in the text.

Virtual idols are no longer a novelty in Korean pop. Groups whose members exist only on screen have topped streaming charts, sold hundreds of thousands of albums and won music-show trophies. Meanwhile, generative tools can now produce whole songs, voices and videos from a text prompt. Together they raise legal questions that Korea, the European Union and Türkiye are answering in different ways: who, if anyone, owns AI-made music; whether a singer's voice is protected when it is imitated rather than copied; and whether listeners are entitled to know what they are hearing.

The earlier version of this article was built around an August 2026 "dataset leak", protest trucks outside agency headquarters, a record-breaking fan boycott and a fast-tracked "AI Music Industry Bill". We have found no record of any of these, and they have been removed. The verifiable record shows a growing range of virtual and AI-generated acts, sharp fan reactions to particular uses of AI, a coalition of Korean music rights bodies demanding consent for AI training, and a copyright society that adopted a rule on AI-assisted songs and withdrew it three weeks later.

From avatars to fully generated acts

Virtual groups with human performers

Two of the best-known virtual groups debuted in early 2023, not in 2024 as the earlier version stated. MAVE:, a four-member virtual girl group, released its debut mini-album on 25 January 2023 through Metaverse Entertainment, a Netmarble affiliate, with Kakao Entertainment co-designing the group, as The Korea Times reported on the day. PLAVE, a five-member boy group under VLAST, debuted in March 2023. According to The Korea Times, each PLAVE member is voiced and performed by a real person whose identity the agency keeps private, and the members take part in writing lyrics and choreography. The group's first mini-album sold more than 200,000 copies in its first week.

By January 2025, The Korea Times was reporting that a PLAVE single had topped Melon's HOT100 and TOP100 charts on its release day and won a music show's top prize. The same report describes other virtual acts: Isegye Idol, created by a YouTuber and active since 2021, and naevis, SM Entertainment's first virtual artist, who debuted in September 2024.

These groups matter legally because, for the most part, they are not "AI idols". A human voice and a human performance sit behind the avatar. Contract law, performers' rights and personality rights therefore attach to identifiable people, even when their names are not public.

Acts generated end to end

A different model has appeared more recently. In April 2026 Dazed reported on GLXE, a trio that debuted in December 2025 with visuals, voices and songs generated by AI, the music made with Suno. The group's creator told the magazine that he writes the lyrics and builds the tracks in the tool, and claimed that about 90 per cent of fans say they do not mind that the group is AI. The same article mentions Syndi8, a short-lived AI girl group introduced by Supertone in 2024, and a fully AI-generated video that SM Entertainment released for an aespa remix in September 2025. It also notes that Suno faces copyright lawsuits from major labels.

None of the sources we reviewed supports the earlier version's claim that a chart-topping virtual group had been trained on scraped voices of real vocalists.

Fan reaction: real, but targeted

We found no record of a mass boycott of AI or virtual idols in Korea. What the record does show is focused reactions to particular products. The clearest case came in June 2025, when RBW, the agency behind Mamamoo and Oh My Girl, offered AI chatbot versions of its artists through Klleon's Almigo app. As Korea JoongAng Daily reported, fans objected that the service exposed artists to harassment, criticised a premium "Spicy" option, and ran a hashtag campaign against the companies. RBW apologised on 6 June and the characters were removed the following day. A fan coalition opposing the app then demanded a fuller explanation, steps to prevent a recurrence and deletion of the development files. Klleon said it had recorded the material itself, with the artists' consent, and denied any link to deepfake technology.

Dazed also describes wider unease among fans about authorship and authenticity, including anger at reports that AI had been used in K-pop videos without disclosure. These reactions are not, on the evidence, a revolt against virtual idols as such. They concern consent, disclosure and the treatment of real artists, which are the same issues the law is now addressing. That is why the word "rebellion" has been dropped from the title.

The industry's answer: consent and registration

A coalition of rights bodies

On 26 February 2026 six Korean music rights organisations launched a joint committee: the Korea Music Copyright Association (KOMCA), the Korea Record Industry Association, the Korea Entertainment Producers Association, the Together Music Copyright Association, the Korea Music Performers Association and the Korea Music Content Association. Their joint declaration, reported by Sports Kyunghyang, calls for a ban on AI training without creators' consent, transparency in AI generation, and a clear institutional distinction between human-created works and AI outputs. KOMCA's president, who chairs the committee, described the next two years as a "golden time" for the survival of the Korean music industry. According to Digital Music News, its priorities include training-data transparency and labels for AI outputs on digital platforms.

This is the "push for consent" in the new title. It is an industry campaign, not legislation. We found no bill before the National Assembly matching the earlier version's description.

KOMCA's registration rule: adopted and withdrawn

KOMCA collects royalties for composers and lyricists, so what it agrees to register decides who is paid. In 2025 it took a strict line. As RouteNote reported in April 2025, it would not register works with any AI involvement, members had to declare that no AI had been used, and if AI use was later confirmed, payments could be suspended and the work deleted.

In early August 2026 KOMCA changed course. Its revised rules allowed registration of AI-assisted works, but only for the parts in which a human had made a substantial, leading creative contribution to the lyrics, composition or arrangement; tracks generated from prompts alone were excluded (Edaily). Kyunghyang Shinmun reported that KOMCA treated "leading" as at least half of the creative input, and that the change followed a Board of Audit and Inspection review which had identified registered works presumed to have been made with generative AI.

The rule lasted about three weeks. On 10 August Rep. Kim Jae-won of the Rebuilding Korea Party said at a press conference that there was no social consensus or legislative conclusion on copyright in AI-assisted music. On 18 August the Ministry of Culture, Sports and Tourism wrote to KOMCA recommending broader discussion across the cultural sector first. On 25 August KOMCA's board withdrew the amendments (Korea JoongAng Daily). KOMCA said that it would begin procedures to cancel registrations already made under the rule, that fully AI-generated works would still not be accepted, and that checks against false filings would continue. Its chair said that setting standards for works that use AI "can no longer be postponed" (Sports Kyunghyang).

The practical stakes are royalty pools. Kyunghyang also reported customer complaints about AI-generated background music in cafés and restaurants, which some venues reportedly play to avoid royalties.

Korean law: what protects a voice?

Copyright requires a human author

Korea's position on AI output is set out mainly in government guidance. The Generative AI Copyright Guide, first published in December 2023 and released in English in April 2024, states that only works expressing human thoughts or emotions can be registered, so AI output without human creative intervention is not eligible, although human creative contributions can be (Ministry of Culture, Sports and Tourism). The guide also advises AI businesses to secure the right to use training data and pay appropriate compensation, and suggests that rights holders who object to training say so in their terms or through technical measures such as robots.txt.

In June 2025 the ministry and the Korea Copyright Commission added a registration guide for works made with generative AI and a guide on preventing disputes over AI outputs (Mondaq). Applicants must separate AI-generated parts from human-made parts, and only the latter are protected. Entering a prompt is generally not enough, and falsely registering AI output as a human work can carry legal penalties.

Copyright does little, however, for a singer whose voice is imitated. A model that produces a new song "in the voice of" an artist need not copy any composition or recording. Performers' rights attach to fixations of an actual performance, not to a resemblance.

Voice as an identity marker

Here Korea has a specific statutory tool. Since 8 June 2022, Article 2(1)(ta) of the Unfair Competition Prevention and Trade Secret Protection Act, inserted by Act No. 18548 of 7 December 2021, has treated as unfair competition the unauthorised use, for one's own business and contrary to fair commercial practice or competitive order, of another person's name, likeness, voice, signature or other identifying marker that is widely recognised in Korea and has economic value (current text; Shin & Kim). The provision names voice expressly.

Its limits are just as clear. It protects people who are already well known, it applies to use in business, and it is framed as a fair-competition rule rather than a free-standing personality right. A cloned voice used to sell a product or a track is the kind of case it was written for; a non-commercial fan cover, or the voice of a trainee or session singer who is not yet widely recognised, may fall outside it.

Labelling under the AI Basic Act

The Framework Act on Artificial Intelligence, passed on 26 December 2024 and in force since 22 January 2026, adds transparency duties. Under Article 31, providers of generative AI must indicate that outputs were generated by AI, and virtual sounds, images or videos that are difficult to distinguish from reality must be clearly notified or labelled; the labelling may be adapted so that it does not hinder artistic or creative expression. Failing to give the advance notice required by Article 31(1), that a product or service runs on generative or high-impact AI, can lead to an administrative fine of up to KRW 30 million under Article 43; the Act sets no fine for the labelling duties themselves (official text; Future of Privacy Forum). This, not an "AI Music Industry Bill", is the real legal basis in Korea for telling listeners that a voice or track is synthetic.

The EU and Türkiye in brief

European Union

The EU AI Act reaches AI music by two routes. On the input side, providers of general-purpose AI models must have a policy to comply with EU copyright law, including identifying and respecting rights reservations made under Article 4(3) of Directive (EU) 2019/790, the text-and-data-mining opt-out, and must publish a sufficiently detailed summary of the content used for training (Article 53(1)(c) and (d)). These duties have applied since 2 August 2025 (Article 113).

On the output side, Article 50(2) requires providers of systems that generate synthetic audio to mark it in a machine-readable format that is detectable as artificial, and Article 50(4) requires deployers of deep fakes, audio included, to disclose that the content has been artificially generated or manipulated. For evidently artistic, creative or fictional works, that duty is limited to disclosing the existence of such content in a way that does not hamper the enjoyment of the work (Article 50). Article 50 applies from 2 August 2026; under the transitional rule now in Article 111(4), providers whose systems were already on the market before that date have until 2 December 2026 to meet the marking duty (Article 111). The AI Act does not itself create a right over one's voice; that is left to national law.

Türkiye

Turkish law reaches the same questions through general rules. Law No. 5846 on Intellectual and Artistic Works (FSEK) defines a work as an intellectual or artistic product bearing the characteristics of its author, and the author as the person who creates it (Article 1/B), which makes copyright in pure machine output difficult to argue. Under Article 80, performers can object to distortion of their fixed performances and hold exclusive rights over the fixation, reproduction and communication to the public of those performances (FSEK). As in Korea, those rights cover actual recordings, not a synthetic imitation.

For the voice itself, the main route is the protection of personality rights in Articles 24 and 25 of the Turkish Civil Code. They allow a person to ask the court to prevent or stop an unlawful attack and to claim damages, as well as the profits made from the attack under the rules on agency without authority (Turkish Civil Code). Where a voice is processed to identify a person, it may also count as biometric data, which Article 6 of the Personal Data Protection Law No. 6698 treats as special-category data whose processing is prohibited unless one of the listed conditions, such as explicit consent, is met (KVKK).

Burhan Doğuş Ayparlar's View

The following is opinion. The verifiable story of AI in K-pop is not about machines replacing idols. It is about consent and disclosure. PLAVE and MAVE: show that audiences accept virtual performers, at least where a human performer stands behind the avatar. The RBW episode shows that fans reject uses of real artists' identities that look imposed on them. The law is moving along the same line.

In some respects Korea's toolkit is ahead of others. It has an unfair-competition rule that names the voice, a labelling duty already in force, and detailed ministry guidance on human authorship. The gap is training consent. The rights bodies' demand for a consent requirement is a campaign, not a rule, and the ministry's guidance still relies largely on rights holders opting out. The unfair-competition provision also protects only those who are already well known. In our view, a narrowly drafted right over one's own voice and likeness, available regardless of fame, would close that gap more cleanly than trying to stretch copyright to cover vocal timbre.

The KOMCA episode carries a second lesson. A collecting society cannot settle authorship questions alone, and the withdrawal was defensible as a matter of process. But the uncertainty it left behind falls on creators who use AI as a tool and now cannot register anything. A public standard, built on the ministry's registration guide, would serve them better than a cycle of private rules and reversals.

For Türkiye, there is no voice-specific rule. Personality rights and data protection law do the work, but they were not designed for synthetic voices and their remedies are slow. The EU and Korean labelling models offer a useful reference point for any future Turkish AI legislation: listeners should be able to know when the voice they hear is not a person's.

Sources

This article is general information and does not constitute legal advice. It reflects the position as of 19 October 2026.