China’s AI openness claims scrutinised in Guardian letter

A letter published in The Guardian challenges China’s claim that its artificial-intelligence ecosystem is fully open, pointing to specific models and licensing restrictions. The critique follows an article by China’s ambassador to the UK, Zheng Zeguang, which praised the country’s openly released AI models.

Zheng’s 30 July 2026 article, titled ‘The future of AI hinges on openness and cooperation. China and Britain can gain much by working together’, highlighted Chinese AI labs behind models such as Qwen, DeepSeek, and Kimi for their role in increasing accessibility. These models have been released openly, benefiting competition and users in the developing world, particularly where hardware is modest.

However, the letter—contributed by Prof Paul H Cleverley and Dr Claire Jenkins—argues that China’s openness is overstated. It cites GeoGPT, a geoscience AI system from Zhejiang Lab, showcased at the July 2026 World AI Conference as an example of jointly governed open science. Despite its promotion as open, GeoGPT does not meet the model openness framework endorsed in a June 2026 UN report, ‘AI Systems as Digital Public Goods’. While GeoGPT releases model weights built on Alibaba’s Qwen, Qwen’s licences are not compliant with Open Source Initiative (OSI) standards. The system also withholds its training data and application source code, and its governance committee reports solely to Zhejiang Lab.

The letter acknowledges the progress made by Chinese AI labs but calls for shared global standards on openness. It suggests that British AI could offer a distinctive path in this regard. On 27 July 2026, China’s Moonshot AI released its most powerful chatbot, Kimi, for free online, further fuelling the debate over accessibility and transparency in AI development.

The authors note that no country’s AI ecosystem is fully open, emphasising the need for clearer benchmarks to assess openness in practice.