China Eyes 3 New AI Controls, Including Criminalizing Model Leaks

In-house counsel at multinational tech firms face a potential new legal minefield as Beijing considers classifying proprietary AI technology as state secrets. The proposals, now under review, would not only criminalize unauthorized transfers of AI models but also limit foreign investment in China’s most innovative AI startups—with penalties that could include economic sanctions and travel bans. Companies that have integrated Chinese AI models into their global products need to reassess their exposure immediately.

Chinese authorities are actively deliberating restrictions on overseas access to the country's most advanced artificial intelligence models, a move that would fundamentally reshape the global AI landscape. Over the past month, officials from the Ministry of Commerce have held closed-door meetings with Alibaba, ByteDance, and AI startup Z.ai to gauge the industry's stance on limiting the export of cutting-edge models, including those still in development. This initiative, revealed by three anonymous sources, marks a significant escalation in Beijing's approach to AI sovereignty, treating large language models and other foundational AI as strategic assets akin to nuclear technology or advanced semiconductors.

“

The timing is especially delicate as open-source AI has blurred traditional lines of control—models like DeepSeek R1 can be downloaded and run locally anywhere, making enforcement of access restrictions a technical and legal challenge.

The proposed restrictions come in the wake of DeepSeek's explosive global success. Its R1 model, released in January 2025, demonstrated that Chinese AI could match or exceed Western performance at a fraction of the cost, sparking a wave of adoption among international developers and enterprises. The popularity of open-source Chinese models has provided a low-cost alternative to proprietary systems from OpenAI and Google, but it also exposed a vulnerability that Beijing now appears intent on closing. By potentially applying the restrictions to future models only, China may be seeking to lock down the next generation while allowing current models to remain accessible, thereby maintaining goodwill while tightening the net.

The discussions reportedly include three key measures: limiting overseas access to both closed-source and more open models, criminalizing the leak or theft of proprietary AI technology under national security law, and introducing new restrictions on foreign investment in domestic AI startups. The severity of the proposals—especially the national security classification—would place AI trade secrets alongside state secrets, subjecting violators to harsh penalties. This mirrors, in some respects, the U.S. approach of viewing advanced AI through a national security lens, but China's potential move goes further by directly controlling the outbound flow of civilian technology that has already permeated global markets.

The implications are profound. For companies like Alibaba, which runs a major cloud business and has invested heavily in its Tongyi Qianwen model series, overseas restrictions could undermine international revenue streams and complicate partnerships with foreign clients. ByteDance, whose Doubao model undergirds many of its domestic applications, might face a strategic dilemma: protect domestic dominance or pursue global ambitions. For Z.ai and other startups, investment curbs could dry up crucial foreign capital, limiting their ability to scale innovation. The global AI market would face increased fragmentation, with developers forced to choose between Chinese models (possibly with restricted access) and Western alternatives, driving up costs and reducing interoperability.