Realtime AI News
Exclusive: Beijing considers curbing overseas access to China's top AI models — Reuters
Beijing is exploring measures to restrict overseas access to China's leading AI models, according to sources cited by Reuters. The potential policy shift could fundamentally reshape the global AI supply chain and open-source ecosystem.

Beijing is considering measures to restrict overseas access to China's top AI models, Reuters reported exclusively on July 10, citing sources familiar with the matter. The news sent ripples through the global AI industry.
The discussions remain at an early stage, with the specific scope and mechanisms still under internal review, the report says. Possible measures include stricter licensing requirements for model exports or limitations on overseas API access to China's high-performance AI models.
The timing of this potential policy shift is significant. Chinese AI models — particularly those from DeepSeek, Alibaba's Qwen, ByteDance, and others — have achieved widespread global adoption in both open-source and closed-source forms. Restricting overseas access would fundamentally reshape the global AI model supply landscape.
For the global open-source AI ecosystem, the implications are profound. Many of China's leading open-source models are widely used by developers worldwide.海外访问若受限, overseas startups and research institutions that depend on China's model ecosystem would face significant challenges in finding alternatives.
This move could also accelerate the fragmentation of the global AI landscape. The US has already restricted high-end chip exports to China. If China responds with reciprocal restrictions on model exports, the global AI supply chain would become increasingly divided.
Key aspects to monitor include the specific policy timeline and scope, whether the restrictions differentiate between open-source models and API access, and how governments and companies worldwide respond.
Why it matters
Beijing's potential move to restrict overseas AI model access could fundamentally alter the global open-source AI ecosystem and model supply chains, accelerating regional bifurcation in AI.
Nearby Updates
All07/10, 16:01
World's first 'embodiment-native' pretrained model LingBot-VA 2.0 goes open-source
Researchers have released LingBot-VA 2.0, claimed to be the world's first 'embodiment-native' pretrained model, and made it fully open-source. The model is designed to give robots a 'brain' that truly understands the physical world from the ground up.
07/10, 15:00
Deutsche Telekom partners with OpenAI to become an AI-native telecom operator
Deutsche Telekom is deepening its partnership with OpenAI to embed AI across customer service, employee workflows, network operations, and voice interactions. Europe's largest telecom operator is pursuing a full AI-native transformation.
07/10, 12:44
Gege AI launches billion-parameter Mandarin music model with ByteDance revenue-sharing deal
Gege AI released a from-scratch trained billion-parameter end-to-end Mandarin music generation model that produces full 3-minute stereo songs with separate vocal and accompaniment tracks. The company also signed a copyright revenue-sharing agreement with ByteDance, enabling AI-generated music to be monetized across Douyin, CapCut, and other platforms.
07/10, 10:56
xAI Unveils Grok 4.5: Coding Agent Costs Slashed 80% with Near-Frontier Speed
xAI has released Grok 4.5, a new model that cuts coding-agent operational costs by approximately 80% while maintaining near-frontier inference speed. However, the model shows higher hallucination rates, suggesting accuracy trade-offs in its design.