DeepSeek, Researchers from Alibaba support China's 'misunderstood' AI regulatory framework

Researchers from the AI start-up DeepSeek and tech giant Alibaba Group have expressed support for China's evolving AI regulatory framework while highlighting the need for clearer feedback mechanisms, according to a paper published in Science last month. The study, titled "China's Emerging Regulation Towards an Open Future for AI", emphasized that China has established an institutional environment favorable to innovation and openness in AI development, but noted that a national AI law could further enhance this system.

The paper aimed to introduce international readers to China's pragmatic approach to AI governance, which has often been misunderstood abroad, explained Zhang Linghan, professor at the China University of Political Science and Law and one of the ten co-authors of the study. Zhang added that China has shifted from following global AI governance trends to leading them, a change reinforced by his work on drafting a proposed national AI law introduced last year.

Joining Zhang were other legal and industry experts, including Fu Hongyu, director of AI governance at Alibaba's AliResearch, and Wu Shaoqing, head of AI governance at DeepSeek. This marked Wu's second recent contribution to AI governance discussions, following his participation in a panel on ethical AI at the Global Open-Source Innovation Meetup in Hangzhou.

The study, led by Zhu Yue, assistant law professor at Tongji University and former ByteDance researcher, noted that China's AI regulatory system is evolving alongside the global popularity of its open-source AI platforms. The framework relies on a combination of AI-related regulations, including pre-deployment filings for AI models and content safety self-assessments.

The authors identified six key pillars of the system, such as exemptions for open-source AI and AI-driven scientific research, efficient judicial handling of AI cases, and phased implementation of new rules. However, they acknowledged gaps, including unclear feedback when filings are rejected and potential risks posed by unregulated open frontier AI models that could be misused if exemptions are too broad.

"Leading Chinese AI firms must be more transparent and evidence-driven in governing frontier models," the authors wrote. Although a unified national AI law has not been enacted, proposed drafts outline liability for AI developers, providers, and users in cases of misuse or incidents. The study also highlighted that both Chinese and international AI companies often lack independent verification mechanisms for frontier AI risk management, while Chinese AI models are increasingly comparable to U.S. models in terms of advanced capabilities.

Fang Liang, head of AI safety and governance at Beijing consultancy Concordia AI, noted that China's governance logic differs from Europe and the United States. In China, AI openness is treated as a safety measure rather than a risk factor.

Author: Sophia Brooks