Mandopop sensation Stefanie Sun has become an internet sensation on Bilibili, China’s largest user-generated video streaming site. However, the newfound interest in Sun, who hasn’t released an album since 2017, is not due to her own musical brilliance. Instead, the viral videos featuring Sun’s voice have been created using artificial intelligence, raising concerns about copyright protection.

In a nostalgic tribute to the golden era of Mandarin pop music, tech-savvy Chinese internet users have employed a deep learning technique called singing voice conversion to mimic Sun’s voice. They have then used this cloned voice to sing a compilation of Mandopop classics. A quick search for “AI孙燕姿” (“AI Stefanie Sun”) on Bilibili reveals hundreds of videos uploaded within the past month. The most popular ones have garnered over one million views. The term’s trending score on WeChat Index, which tracks keywords across the social and content ecosystem of the super app, skyrocketed to 50,000 on May 5, just two days after being at zero.

The fascination with deepfake singing is not limited to China. In late April, an AI-generated song featuring the voices of The Weeknd and Drake gained significant popularity on Spotify and TikTok. Grimes also generated buzz by announcing that she would share 50% of the royalties from any successful AI-generated song using her voice.

Like in the West, current copyright laws in China do not specifically address AI-generated songs that exploit celebrity voices. However, Chinese regulators have been swift in formulating legal restrictions on the broader field of synthetic technology. As Chinese tech companies raced to leverage advances in generative AI with products like Midjourney and ChatGPT, China enacted a regulation in November to establish guidelines for the responsible use of cutting-edge technology. Service providers are required to verify users’ identities and maintain records of any illegal activities.

While Sun’s deepfake videos gained traction on Bilibili, a hub for mashup content, they have also been shared on other major social media platforms, including Douyin, the Chinese version of TikTok. Douyin has been proactive in addressing the legal implications of the proliferation of AI-generated content. The ByteDance-owned company recently published guidelines for AI-generated content, largely based on China’s new regulations on synthetic technology. The guidelines state that content creators must label AI-generated content accordingly and assume responsibility for any consequences arising from such content. The platform strictly prohibits any content that infringes on copyrights and promises “severe punishment” for offenders.

The crucial question remains whether songs created using tools that imitate singers’ voices without their consent violate the artists’ rights. Sun has not publicly responded to the dozens of songs produced using her AI voice.

Generative AI has found widespread application in addressing emotional voids, whether by helping people remember deceased loved ones or, in the case of the deepfake Sun, filling the gap left by the lack of quality Mandopop music today. As one AI product manager tweeted, “It’s as if Sun’s fans have suddenly entered festival mode.”