China’s approach to AI regulation is under scrutiny, with experts questioning whether the country prioritizes development speed over safety. While China has implemented strict AI regulations, some argue that these rules focus more on content and public-facing applications than on advanced models’ safety.
China’s AI regulations: strict but selective
According to The Economist, China has adopted a security framework identifying 54 risks across three categories: unintended model behavior, criminal use, and societal impact. However, SemiAnalysis notes that China’s regulations primarily target AI-generated content and public applications, such as requiring warnings for AI-created content and implementing rules for virtual companions.
In contrast, China does not impose special obligations on its advanced models, like those from DeepSeek, Alibaba, or Moonshot. The European Union’s Artificial Intelligence Act and California’s Transparency in Frontier Artificial Intelligence Act of 2025 take a broader approach, addressing advanced models directly by mandating safety assessments and transparency measures.
Safety evaluations and censorship concerns
A SemiAnalysis study revealed that only 3.6% of models launched by China’s top developers published safety evaluations. This finding raises concerns about the effectiveness of China’s regulations in ensuring AI safety. Additionally, Jordi Calvet-Bademunt and Jacob Mchangama from Vanderbilt University highlight that Chinese regulations require AI providers to adhere to socialist values, effectively embedding censorship into the system.
The Economist admits that many Chinese regulations aim to prevent AI models from contradicting the Communist Party on sensitive issues, meaning these platforms lack transparency and suffer from serious biases.
Thus, China’s efforts in this area are directed more toward censorship than safety.
Leave a Reply