Beijing meets Alibaba & ByteDance, mulls curbing overseas access to China's top AI models- Here's why

Chinese authorities have held meetings with top tech firms over the past month about potentially restricting overseas access to China's most advanced AI models, including those yet to be released, according to a Reuters exclusive report.

If adopted, the measures would mark one of China's strongest moves yet to treat advanced AI as a strategic national asset, placing it alongside semiconductors and other sensitive technologies that are increasingly subject to government oversight.

Why Beijing wants tighter control over China's AI models

China's AI industry has expanded rapidly over the past year, with domestic companies emerging as serious competitors to US firms.

The momentum accelerated after DeepSeek released its R1 reasoning model, demonstrating that Chinese developers could build powerful AI systems at substantially lower costs.

Since then, companies including Alibaba and ByteDance have rolled out increasingly capable models, while startup Z.ai has drawn global attention after its GLM-5.2 model demonstrated performance approaching leading US systems at a fraction of the price.

Officials from China's Ministry of Commerce have reportedly spent the past month consulting major AI developers on possible safeguards for the country's most advanced technologies, according to three people familiar with the discussions cited by Reuters.

Alibaba, ByteDance and Z.ai among companies involved

The consultations reportedly included representatives from Alibaba, ByteDance and Z.ai, all of which develop advanced AI models using different deployment approaches.

Alibaba's Qwen and ByteDance's Doubao have become two of China's most widely used AI platforms, while Z.ai has emerged as one of the country's fastest-growing AI startups.

According to Reuters, officials discussed whether restrictions should apply to both closed-source models and more open-weight systems that developers can download, modify and deploy independently.

AI theft could face tougher legal penalties

The discussions extended beyond export restrictions.

According to Reuters, officials examined whether the theft or unauthorised disclosure of proprietary AI technology should become an offence under China's national security law.

Authorities also considered tightening oversight of investment into domestic AI startups by introducing additional restrictions on who can fund emerging AI companies.

China, US increasingly mirror each other's AI security concerns

The Trump administration has also introduced restrictions on access to advanced AI models. In June, Washington barred foreign nationals from using Anthropic's most advanced Fable and Mythos models over concerns that the technology could be exploited by military or intelligence organisations in countries including China and Russia.

Anthropic temporarily disabled access worldwide because it could not verify users' nationalities in real time. While export controls on Fable have since been lifted following additional safeguards, Mythos remains available only to selected trusted US organisations.

Some American AI researchers have also argued that US should regulate access to advanced Chinese AI models.

Beijing reportedly concerned about Anthropic's Mythos model

According to Reuters, Chinese officials have expressed concern that Anthropic's cybersecurity-focused Mythos model could identify software vulnerabilities that might be used against Chinese interests.

China has steadily expanded oversight of AI this year

Earlier this year, Chinese authorities investigated several domestic AI startups that had expanded overseas to determine whether export control laws had been breached, Reuters reported.

In April, China's state planning agency also ordered Meta to unwind its proposed $2 billion acquisition of Chinese-founded AI startup Manus. The following month, Beijing introduced broader rules tightening scrutiny of overseas transactions involving Chinese technology, data and national security.