Chinese AI developers such as DeepSeek and Xiaomi release updated “open-weight” models every month, allowing users to take the core software and run it without paying an expensive ongoing licence. Writing in the Observer, Phillip Inman argues that Britain should reject the lure of this cheap technology.
The open-weight model and its risks
Inman compares cheap Chinese AI to the debate over Chinese electric cars in Britain, which are cheaper than other EVs sold in the UK. He writes that Beijing subsidises Chinese EVs deliberately to boost exports and undermine foreign domestic car industries, and argues cheap AI takes this export threat to a higher level.
Ross Burley, director of the Centre for Information Resilience (CIR), an independent non-profit group that exposes human rights abuses and counters disinformation, said Chinese AI comes with extra baggage. He said DeepSeek cannot access the properly isolated use of its software, “but that addresses just part of the risk. Running a model on your own servers doesn’t automatically remove any political censorship. Or indeed repair weaknesses in any security defences.”
An evaluation report by the US National Institute of Standards and Technology (Nist) found political censorship in downloaded DeepSeek models running on its own infrastructure, meaning hosting those models in the UK would not by itself remove censorship. “We should be deeply concerned about embedding Beijing’s political restrictions into systems that businesses and public services rely on for information and decisions,” Burley said.
Data collection and state access
According to Inman, the policy governing DeepSeek’s online chatbot says it collects prompts and uploaded material, processes and stores personal data in China, and permits disclosure to Chinese authorities.
Michael Wooldridge, a professor of AI at the University of Oxford, says it is not unreasonable to assume data inputted into the chatbot could be shared with the Chinese state.
Choice between Chinese and US systems
Inman writes that users face a choice between a cheap Chinese open system they can tweak for their own ends or expensive US software with features they probably do not need. He says both types have the potential to destroy humanity, but in day-to-day use only one comes with at least some effort to provide security and a benign intent — and it is not the Chinese option.
He notes that Chinese developers launch AI models by the dozen each month at a fraction of the cost of US equivalents. Some tech analysts say Chinese models add to the rich tapestry of AI, supplementing the closed, proprietary models of OpenAI and Anthropic, and that European and Indian versions are in development. Meta is also developing open AI models but, like other challengers, lags in the race to win over customers.
Consequences for policy
Inman argues the consequences for policymaking are profound. He writes that the UK should reject the lure of cheap Chinese AI and seek to better tame US AI while supporting domestic and EU suppliers. Relying on Chinese tech, from cars to AI, he says, will hollow out industry and undermine the security of our data.
