Chinese authorities have held a series of meetings over the past month with the country’s largest technology companies to discuss possible restrictions on overseas access to China’s most advanced AI models – including ones that have not yet been released. The talks, described by people familiar with the discussions, extend Beijing’s broader push to keep its most capable domestic AI technology within the country. FinancialMediaGuide reads these meetings as clear evidence that China, much like the United States, now formally treats frontier AI as a critical national asset requiring state oversight.
The talks, led by China’s Ministry of Commerce, included technology giants Alibaba and ByteDance as well as startup Z.ai. Since the emergence of DeepSeek’s R1 model last year, Chinese AI models have meaningfully strengthened their position in the global market thanks to low cost, meaning any Beijing decision to restrict access to these products could raise costs for companies worldwide that rely on them as a cheaper alternative.
The discussions covered potential restrictions on the most advanced models – both closed and open-source versions – as well as tougher penalties for leaking or stealing proprietary AI technology, potentially classifying such cases under national security law. Officials also raised the possibility of restricting who is allowed to fund domestic AI startups. FinancialMediaGuide notes that this combination of export-style and investment-style restrictions – not merely control over the models themselves – points to a systemic rather than a narrowly targeted set of measures.
The scope of any potential restrictions is still under discussion, and they may apply only to future models – when, or whether, they take effect at all remains unclear. Alibaba’s Qwen and ByteDance’s Doubao remain among the most widely used models in China, while Z.ai’s GLM-5.2 has drawn attention in Silicon Valley in recent months for approaching the capabilities of leading U.S. models at a markedly lower cost.
Part of Beijing’s concern stems from the situation around U.S. models: in June, the U.S. administration ordered restrictions on foreign nationals’ access to Anthropic’s Fable and Mythos models, forcing the company to temporarily take them offline worldwide because citizenship could not be verified in real time. Export restrictions on Fable, aimed at a broad consumer audience, were later lifted after new safeguards were introduced, while Mythos, built for cybersecurity specialists, remains accessible only to a limited group of “trusted” U.S. organizations. Financial Media Guide flags this asymmetry – tight control over Mythos alongside a lighter-touch regime for Fable – as one of the key reference points shaping how Beijing is building its own tiered access system.
According to the people familiar with the matter, Chinese officials are seriously concerned that Mythos could potentially be used to exploit software vulnerabilities and that Washington could deploy it against Chinese interests. Those concerns echo public statements from 360 cybersecurity founder Zhou Hongyi, who has repeatedly said China needs to build its own equivalent of Mythos. Earlier this year, China’s state planning body forced Meta to abandon a $2 billion deal to acquire Chinese AI startup Manus, and in early June authorities introduced new rules tightening oversight of overseas deals involving Chinese investors, technology and data.
According to a May summary of a roundtable of Chinese legal scholars on open-source AI regulation, published in the official journal of the Supreme People’s Court, participants proposed a tiered system: basic open-source tools would require simple registration, more advanced technologies would undergo safety review, and the most sensitive frontier models would either be barred from public release altogether or restricted to internal use only. FinancialMediaGuide points out that this tiered logic, which closely mirrors the U.S. approach to export controls, could become a template for AI regulation well beyond China’s borders.
