Report: Beijing considering restricting access to China's leading AI models

Contents

AI as a matter of national security. In June, Chinese authorities held consultations with leading domestic technology companies about a possible restriction of overseas access to China's most advanced AI models. Reuters reported on Tuesday, citing three people familiar with the discussions.

The discussions, which included representatives from Alibaba, ByteDance, and the AI startup Zhipu AI (known as Z.ai), underscored, according to Reuters, that Beijing, similar to Washington, now considers cutting-edge artificial intelligence (AI) a critical national asset requiring state control. The meetings, led by the Chinese Ministry of Commerce, focused on restrictions for China's most advanced AI models – both closed-source and more open versions, according to two of Reuters' sources.

Theft of AI technology as a criminal offense

All three companies mentioned have various AI models. Some are closed-source, while others are open-weight models, meaning users can download, run, and adapt the underlying systems. Among the most widely used AI models in China are Alibaba's Qwen and ByteDance's Doubao. Z.ai recently caused a stir in Silicon Valley as the performance of its GLM-5.2 model approaches that of leading US models like Anthropic's Mythos – and at a fraction of the cost.

According to one of the sources, the meetings also discussed including the leakage or theft of proprietary AI technology as a criminal offense in the national security law. Furthermore, government representatives discussed possible new regulations on who is allowed to finance domestic AI startups.

The scope of possible restrictions is still being discussed, according to Reuters. Two of the sources suggest that they might only apply to future AI models. It is also unclear when or if these measures will come into effect at all. Neither the Chinese government nor Alibaba, ByteDance, or Z.ai responded to Reuters' inquiries on the matter.

Danger to national security?

The increasingly powerful AI models because of their potential for misuse, have raised national security concerns not only in Beijing. When the US AI company Anthropic unveiled its new powerful AI model Mythos in April, capable of identifying security vulnerabilities in software systems, there was considerable excitement. Mythos is too powerful and therefore too dangerous for the public, Anthropic decreed. In mid-June, the US government demanded a ban on Anthropic's AI models Fable and Mythos. Foreign nationals should not have access to Anthropic's most advanced AI models, it was stated. Fear of espionage by China is said to be the trigger for the bans. Later, the US government partially lifted the export restrictions.

According to two of Reuters' sources, Chinese authorities, in turn, are deeply concerned that Mythos could exploit software vulnerabilities and that the US government could use the model against Chinese interests. Just recently, Alibaba banned Anthropic's AI model Claude. Due to fear of surveillance, Alibaba employees are no longer allowed to use Claude.

China has also taken numerous measures this year to retain AI technology developed domestically within the country. At the end of April, the government vetoed the two billion US dollar acquisition of the Chinese AI startup Manus by the Facebook group Meta. As part of the review of the deal, Chinese authorities prohibited Manus CEO Xiao Hong and chief researcher Ji Yichao from leaving the country. At the end of May, the government expanded existing exit restrictions for AI talent, including AI executives from private companies like Alibaba and DeepSeek. As several Reuters sources report, Beijing also initiated investigations this year against Manus and other Chinese AI startups that have emigrated abroad to determine if they have violated export control laws.

New regulations and possible tiered model

A few weeks ago, Chinese authorities issued far-reaching new regulations that tighten control over foreign business involving Chinese investors, technology, data, and national security. These could now be further tightened – as suggested by the consultations between the government and companies.

However, Reuters was unable to ascertain how possible new restrictions on overseas access to Chinese AI models might specifically look. In this context, the news agency refers to a discussion round held in May by Chinese legal experts on regulations for open-source AI. According to a summary of the discussions published in an official journal of the Supreme People's Court, participants proposed a multi-tiered system: simple open-source AI models would only be subject to a straightforward notification requirement, advanced models would undergo security reviews, and top-tier models would either not be made publicly available or only used domestically.

Since the introduction of DeepSeek's R1 model last year, Chinese AI models have gained significant global importance thanks to low costs and rapidly growing performance. If the Chinese government decides to significantly restrict access to Chinese AI models, this could impact AI markets, as it would likely involve rising costs for many companies, according to Reuters.

(akn)