AI "Steals" Voices, Famous Voice Actors File Claims One After Another! AI "Face-Swapping" Ruled Against, How to Solve the "Voice Theft" Dilemma? Lawyers Point Out the Predicament: Difficult Proof, High Costs, Low Gains

With the rapid advancement of AI (artificial intelligence) technology, actors' faces and voices have become prime targets for "poaching."

In 2025, a short drama suspected of using AI face-swapping to replace Dilraba Dilmurat drew industry attention. A year later, the Beijing Internet Court announced the related portrait rights dispute case, clarifying the infringement responsibilities of the production and broadcasting parties, thereby defining the judicial boundaries for portrait rights protection. Meanwhile, well-known voice actors Bian Jiang, Ji Guanlin, and others have repeatedly spoken out to defend their rights, directly pointing to AI "stealing" their voices.

Under the swift development of AI technology, the "faces" and "voices" of artists have become resources that are repeatedly stolen. "Faces can be recognized at a glance, but voices are difficult to judge accurately, making evidence collection and rights protection costly and less beneficial," said Jia Yuqian, a lawyer at Long'an Law Firm in Shanghai, in an interview with the Daily Economic News. From "visible" face-swapping to "audible" voice theft, AI technology is challenging the boundaries of actor personality rights with increasing difficulty.

"Resembling but not identical" also constitutes infringement

In April last year, a female actor in a short drama was suspected of having her face replaced with Dilraba Dilmurat's using AI technology, sparking controversy. On March 20 this year, the Beijing Internet Court announced the trial results, ruling that the short drama's production team used AI face-swapping technology to splice the plaintiff's portrait onto the character's face, causing the public to mistakenly believe she participated, which constituted infringement; the broadcaster, as the platform, failed to fulfill its reasonable review obligations, also infringing on the plaintiff's portrait rights. Ultimately, both defendants were ordered to apologize and compensate for economic losses.

The case involved a total of 44 episodes, approximately 90 minutes long, with two segments using AI face-swapping technology. The court held that although AI-generated portraits differ from the actual person, as long as they can be recognized by the general public or specific industry groups, it can be deemed as using a natural person's portrait.

Deng Yile, a lawyer at Beijing Xingquan Law Firm, told the Daily Economic News that this case sets a judicial bottom line that "AI creation must not infringe on personal rights." Its significance lies in clarifying that AI face-swapping that is "resembling but not identical" also constitutes infringement. It emphasizes that "technological neutrality does not mean exemption from responsibility," closing the loophole of evading infringement liability through AI technology based on evidence rules.

The reporter noted that while the landmark ruling on AI "face-swapping" rights protection has been made, another key aspect of the film and television industry—voice acting—is also facing AI "voice theft." Recently, Ye Qing, who voiced Li Xiaojun in "Sweet Honey," and Liu Jianming in "Infernal Affairs," discovered that her voice had been highly reconstructed and used in an AI-generated short drama. The well-known voice actor Ji Guanlin's voice was also imitated and used in a "modified" Indian version of "Zhen Huan Zhu," prompting collective rights protection.

Sources told the reporter that AI-generated voices are not only used to produce modified short dramas and fake advertisements but may also be employed in scams, posing significant legal risks and reputational damage to voice actors.

Not only individual voice actors but also leading domestic voice companies like Yin Xiong Lian Meng have publicly opposed AI "voice stealing." They issued a statement saying, "Voice is the fundamental livelihood of voice actors and, like appearance, has exclusive personality rights protected by relevant laws and regulations." They explicitly oppose any individual, organization, or platform collecting, recording, or extracting voice materials of their voice actors without written authorization and using them for AI training, speech synthesis, voice replication, and other purposes.

Why is AI "voice theft" rights protection so difficult?

From the perspective of the film and television market and industry, Jia Yuqian believes that AI face and voice theft issues can cause audience skepticism and disgust, reducing industry credibility; they may also lead to mutual imitation among industry players and exacerbate technological abuse. "The legal risks are obvious. The temptation of low costs can lead to misuse of technology, resulting in vicious competition within the industry, squeezing out high-quality works, and creating a market phenomenon where 'bad money drives out good.'"

Compared to AI "face-swapping," rights protection for AI "voice theft" is more challenging, not because there is no legal basis, but because practically, the threshold is much higher. According to the Civil Code, protection of natural persons' voices refers to the relevant provisions for portrait rights.

In 2024, the Beijing Internet Court clarified in the country's first AI-generated voice personality rights infringement case that, under recognizable conditions, the scope of natural persons' voice rights can extend to AI-generated voices. However, "having a basis" does not mean "easy to defend." Jia Yuqian pointed out that "faces can be recognized at a glance, but voices are difficult to judge precisely."

She further explained that although voices are protected by law similar to portraits, the path to rights protection faces three major difficulties: first, evidence collection is hard—facial recognition is much easier than voice recognition, and ordinary voice actors find it difficult to prove that AI training data came from themselves; second, the costs are high and the benefits small—star teams have professional legal support, but ordinary voice actors cannot afford it; third, the civil litigation principle of "no action, no response" means voice actors must proactively sue, which circles back to the first two issues.

Deng Yile also added to the reporter that voices are "transient" and difficult to identify infringement immediately, and retrieval is challenging. Currently, it mainly relies on manual listening, which is highly inefficient. Additionally, voices can be edited, sped up, or spliced, making the original state hard to fix, further increasing the difficulty of rights protection. While facial portrait rights have legal "shields," voice personality rights face the awkward situation of "no clear regulations, no unified standards for identification, and no efficient channels for rights enforcement."

Full-chain regulation and technological watermarks may be key to breaking the deadlock

Whether it is AI face-swapping or "voice theft," both reflect some practitioners' opportunistic mindset amid technological waves. An industry insider told the Daily Economic News that, amid increasingly fierce competition in the short drama market, some production companies believe in "traffic first" and resort to "cutting corners" by "planting" stars' faces or voices into works to attract attention. "This disrupts the industry and market, and we strongly oppose such opportunistic practices."

Several industry and legal experts pointed out that addressing the challenges brought by AI technology requires joint efforts to defend actors' "face and voice."

Deng Yile called for the improvement of comprehensive regulation of AI deep synthesis. She noted that the difficulty in curbing voice infringement lies in the opacity of training data and unregistered model use. She suggested establishing informed consent rules for data collection, implementing registration and labeling obligations for commercial AI tools, and controlling "upstream" to ease the burden of "downstream" accountability. Additionally, establishing smooth complaint and feedback mechanisms, along with rapid takedown and traceability systems, is crucial.

At both industry and individual levels, combining preemptive measures with post-incident rights protection is especially important. Jia Yuqian recommended that actors include clauses in contracts with partners specifying that their voices cannot be used for AI training, synthesis, or cloning without authorization. Embedding watermarks or other technical means in commercial voice work can also help trace and hold infringers accountable.

Meanwhile, Jia Yuqian also pointed out that besides litigation, the fastest and lowest-cost method of rights protection is platform cooperation—promptly reporting and removing infringing content. Industry collective rights actions can also amplify influence and reduce individual costs. Coupled with administrative complaints and public statements declaring infringement, these measures can support rights enforcement.

Reporter | Song Xinyue

Editor | Chen Kemin, Yang Jun, Du Hengfeng

Proofreader | Cheng Peng

****** | Daily Economic News nbdnews Original Article |**

Unauthorized reproduction, excerpting, copying, or mirroring is prohibited.