A pair of letters published in the Guardian has pushed back on claims made by China's ambassador to the UK that Chinese artificial intelligence development is a model of openness. The letters were written in response to an opinion piece by Ambassador Zheng Zeguang, published in the same newspaper on 30 July, which argued that AI's future depends on openness and cooperation between China and Britain.
In his article, Zheng pointed to Chinese leadership on openly released AI models as evidence of Beijing's commitment to shared progress, and proposed deeper UK-China collaboration on standards and governance. The ambassador's piece arrived as Chinese AI labs have drawn global attention for releasing powerful models at little or no cost to developers worldwide.
Open weights are not the same as open source
The letter writers, Prof Paul H Cleverley and Dr Claire Jenkins, acknowledge that Chinese labs behind Qwen, DeepSeek and Kimi have released competitive models that can run on modest hardware, a genuine benefit for users in the developing world. But they contest the idea that this amounts to true openness.
The distinction matters technically. What Chinese labs typically publish are model weights, the numerical parameters that determine how a model behaves, which lets developers download and run the software freely. According to Foreign Policy, that open-weight approach lets users download the model and use it however they wish, unlike closed systems from firms such as OpenAI and Anthropic that only work through a subscription. What is usually withheld, by Chinese and Western developers alike, is the training data, the underlying code and the methodology used to build and test the model, the elements that would allow independent verification of how a system actually works and what biases or risks it might carry.
“"If it is closed weight, that means that functionally the only way you can access the model is through the company providing it," one AI policy analyst told Foreign Policy, describing the contrast with open-weight systems that users can run entirely on their own.”
A dispute that stretches beyond China
The letters land amid a live international argument over what to do with these Chinese models. In Washington, officials have reportedly weighed restrictions on their use over surveillance and security concerns, even as Nvidia chief executive Jensen Huang has publicly praised Chinese open models such as Kimi and DeepSeek as excellent and argued they should be embraced rather than banned, according to Fortune. That split, welcomed by some as a boon for cost-conscious developers and distrusted by others as a security risk, underscores how contested the meaning of "openness" has become in AI policy circles.
For British and European readers, the exchange carries particular weight. The UK has positioned itself as a hub for AI safety research since hosting the 2023 AI Safety Summit at Bletchley Park, and British ministers have repeatedly framed transparency and independent evaluation as central to responsible AI development. If neither Chinese nor American labs are disclosing training data or testing methods in full, the letter writers imply, Britain has an opening to define what genuine openness should look like rather than accepting the framing offered by either side.
A distinctive path for British AI
Dr Claire Jenkins, writing separately, argues that Britain need not choose between the American model of costly, closed commercial systems and China's open-weight but data-opaque approach. Instead, she suggests the UK could carve out a distinctive path by insisting on shared, verifiable openness standards, covering not just model weights but training data provenance and safety testing, as a condition for international AI cooperation.
The wider point of both letters is a call for consistency. If openness is going to be a selling point in diplomatic pitches, whether from Beijing, Washington or London, it should mean the same rigorous thing every time it is invoked, rather than shifting depending on who benefits from the claim.
This article is free to read. It always will be — no paywall, no account, no tracking.
