China Considers Restrictions on Global Access to AI Innovations

In recent discussions, Chinese officials have been engaging with major tech firms concerning the imposition of limitations on foreign access to the country’s most sophisticated artificial intelligence (AI) models. This includes models that are yet to be made public.

Sources familiar with these meetings indicate that representatives from companies such as Alibaba, ByteDance, and AI startup Z.ai participated in the conversations. These insiders requested anonymity due to the sensitive nature of the discussions.

The talks are part of China's broader strategy to safeguard its leading AI technologies, reinforcing the notion that advanced AI capabilities are vital national assets in need of stringent protection, akin to U.S. measures.

Future AI Model Limitations Under Review

Led by the Ministry of Commerce, the meetings explored the likelihood of implementing restrictions on the most advanced AI models produced in China. The proposed measures would encompass both proprietary and more accessible models, as revealed by multiple sources.

While the discussions proceed, the precise nature of any restrictions remains uncertain, with indications that new rules might only pertain to upcoming AI developments, and the timeline for any implementation is still unclear.

Neither the Ministry of Commerce nor the National Development and Reform Commission responded to requests for comments from Reuters.

Alibaba, ByteDance, and Z.ai similarly refrained from offering any remarks.

Stricter Penalties for AI Technology Breaches Proposed

In these discussions, officials also considered classifying the theft or unauthorized dissemination of proprietary AI technologies as a violation of China’s national security laws, according to one insider.

Moreover, there were suggestions about the need for new rules governing financial support for local AI startups.

Specifics on these proposals have yet to be determined, and no conclusion has been reached.

Rapid Global Expansion of Chinese AI Models

Since the launch of DeepSeek's R1 model last year, Chinese AI offerings have surged in global markets, benefiting from their competitive pricing and operational efficacy.

Market analysts suggest that if China restricts foreign access to these advanced AI frameworks, international businesses could see increased costs associated with AI due to limited availability of cost-effective options.

Among the leading models in China are Alibaba’s Qwen and ByteDance’s Doubao.

Start-up Z.ai has gained attention in Silicon Valley for its GLM-5.2 model, which demonstrates near parity with top U.S. models at a significantly lower cost.

US AI Security Approaches Reflect Similar Concerns

The discussions taking place in China coincide with the United States' increasing vigilance over advanced AI technologies.

The Trump administration has raised alarms about the potential for advanced American AI systems to be utilized by military or intelligence agencies in nations such as China and Russia.

In a recent initiative, the U.S. limited foreign access to Anthropic’s cutting-edge AI models, Fable and Mythos. Given difficulties in verifying user nationalities, access was temporarily suspended for all users globally.

After implementing additional safeguards, controls on Fable were lifted; however, Mythos remains restricted to selected trustworthy organizations within the U.S., where it serves cybersecurity professionals.

Some experts in U.S. AI circles have argued for regulation of Chinese AI products.

Concerns Arise Over Anthropic's Mythos Model

Notably, Chinese officials have expressed particular worries regarding Anthropic’s Mythos model, which could uncover software vulnerabilities and might be leveraged against Chinese interests, according to two sources.

Such concerns have also been voiced in Chinese state media and by Zhou Hongyi, founder of cybersecurity firm 360, advocating for the development of a comparable AI model in China.

Strengthening AI Regulatory Frameworks in China

This year, China has initiated various measures aimed at tightening regulation over its AI sector.

In April, state authorities mandated Meta to abandon its $2 billion acquisition of the Chinese-founded AI firm Manus.

Additionally, in early June, China enacted regulations intensifying control over overseas transactions concerning Chinese investments, technology, data, and national security implications.

Investigations into Manus and other Chinese AI startups that have moved abroad are underway to assess compliance with export regulations, as per two insiders and an additional source.

Manus has not responded to requests for comment from Reuters.

Proposed Multi-Tier System for AI Models

While it remains uncertain how any forthcoming restrictions on international access to Chinese AI models would be enforced, these topics were broached during a roundtable held in May with Chinese legal experts concerning open-source AI regulations.

According to an overview published by the Supreme People’s Court’s official journal, experts recommended a multi-tier system that would categorize AI models based on their sensitivity.

Under this framework, basic open-source AI tools would require minimal filing, while more sophisticated technologies would undergo thorough security reviews, potentially prohibiting the release of the most sensitive AI models to the public or restricting them for domestic use only.