Source: theguardian.com
The price is only part of the choice
The appeal is straightforward: adapt an open model to a specific task, rather than pay for a more expensive American product with features a user may not need. Chinese models cost a fraction of their American counterparts, and some technology analysts see them as adding diversity alongside closed systems from OpenAI and Anthropic. European and Indian models are also being developed; Meta is expanding its open models but still trails competitors in winning customers.
The article’s concern is not simply where a model is hosted. Ross Burley, head of the Centre for Information Resilience (CIR), an independent nonprofit that works on human-rights abuses and disinformation, says DeepSeek does not allow users to fully isolate their use of its software. He also warns that local hosting alone does not resolve political censorship or security weaknesses.
A US National Institute of Standards and Technology (NIST) assessment found political censorship in downloaded DeepSeek models running on their users’ own infrastructure. That makes server location an incomplete answer to the question of what a model will say.
DeepSeek’s chatbot policy says the service collects prompts and uploaded material, processes and stores personal data in China, and may transfer it to Chinese authorities. Michael Wooldridge, professor of AI at the University of Oxford, says it is reasonable to assume that information entered into the chatbot could be passed to the Chinese state.
The risk of dependence
The case against treating cheap Chinese technology as an uncomplicated bargain extends beyond AI. The article compares the appeal of Chinese electric vehicles in Britain with the concern that Beijing subsidises them to increase exports and weaken other countries’ car industries. In AI, the potential dependence reaches into the systems businesses and public services use to make decisions and handle information.
The article presents Beijing’s use of AI at home as part of a wider political concern: the state can use the technology to identify people in images and track their activity. It also contrasts China’s approach abroad with Russia’s use of AI to intensify cyberattacks on European institutions and companies, arguing that China’s aim is to expand reliance on its products and technology.
For British policy, the proposed answer is to be more cautious about Chinese AI, scrutinise American development more closely, and support suppliers in Britain and the EU. My concern is that the argument risks making model origin stand in for a more specific test: what controls, data practices and security weaknesses apply to each system? The evidence cited here gives a reason not to assume that local hosting is enough, but not a comparison of those risks across the available models.
That gap matters when the cheaper option is also the one buyers can customise. Without clearer distinctions between a model’s price, its deployment terms and its behaviour, procurement could turn a cost-saving decision into a dependency decision—one that is harder to reverse once the technology is embedded in essential services.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X