Artificial intelligence Photo: VCG

During this MKsportsyear's National Two Sessions, Lü Hongbing, a member of the 14th National Committee of the Chinese People's Political Consultative Conference (CPPCC), the country's top political advisory body, who is also supervisor general of the All-China Lawyers Association, has focused on legislative issues related to generative artificial intelligence (AI).

On Wednesday, in an interview with Global Times, Lü stated, "In recent years, AI technology has developed rapidly. Following the sudden emergence of DeepSeek, some have even predicted that 2025 will mark the beginning of the AI terminal application era. However, as real life is already complex, the addition of AI tools makes it even more intricate. Under such circumstances, accelerating the establishment of administrative regulations on generative AI is imperative."

Lü highlighted that generative AI has grown rapidly and gained attention. "As of now, nearly 200 large-scale generative AI models have been registered and approved for public service in China, with over 600 million registered users. AI is reshaping work habits and lifestyles," he noted.

As a practicing lawyer, Lü and his colleagues have encountered multiple legal cases involving generative AI. "For instance, families of the deceased questioned whether AI-generated videos by former employers violate portrait, reputation, or privacy rights. In another case, AI-generated avatars and voices of 'old classmates' were used in deepfake scams — raising questions about legal recourse?" Lü noted.

Moreover, Lü pointed out issues like AI chatbots engaging in inappropriate, borderline, or violent conversations, harming minors' development. He also noted that AI-generated violent images have circulated online. "Given these challenges, China should accelerate the introduction of administrative regulations on generative AI," he stressed.

Lü analyzed that, compared to traditional legislation, AI-related laws face unique difficulties due to the uncertainty of rapidly evolving technology, unpredictable industry shifts, and unforeseeable risks. He suggested that while conducting thorough research, learning from international experiences, and advancing the legal process, China could first focus on a specific area—generative AI—and adopt a "small, fast, and agile" legislative approach to swiftly introduce administrative regulations.

"The core of these regulations should be the integration of socialist core values across the generative AI ecosystem," Lü noted. "Socialist core values serve as a legal principle in China. Purifying the generative AI ecosystem requires not only moral and ethical constraints but also legal guidance based on these values. This approach ensures stronger, more sustainable governance that stabilizes expectations and benefits long-term development."

Lü elaborated that the proposed administrative regulations should clearly define the rights, obligations, and responsibilities of different stakeholders, including AI developers, service providers, users, regulators, and the public.

For example, AI developers should rigorously filter and classify training data to ensure the legality and appropriateness of data sources. They should eliminate inappropriate content from existing models through techniques like data erasure or retraining with cleaner datasets. 

Additionally, manual review mechanisms should be implemented to reassess flagged content. Developers should also promote open research, community collaboration, and whistleblower mechanisms to optimize AI model filtering systems.

A key issue in AI-generated content is copyright protection. Lü pointed out two critical aspects: whether training on copyrighted material constitutes infringement, and the legal status of AI-generated works and their authorship. Internationally, data mining and training generally enjoy relaxed copyright restrictions as long as they remain non-commercial. However, once AI-generated content enters the commercial domain, strict copyright protection applies.

From a legal development perspective, requiring explicit permission for every dataset used in training is impractical, and could hinder new technological innovations. Instead, the law should emphasize usage rights over ownership, ensuring fair compensation for commercial applications. Lü proposed that AI developers should be required to pay royalties for commercial usage and negotiate agreements with rights holders.