British businesses and public services are facing a critical infrastructure decision as Chinese AI developers like DeepSeek and Xiaomi flood the market with open-weight models that offer core software without expensive ongoing licenses. While the price tag is attractive, a new analysis published in The Guardian argues that these tools introduce severe geopolitical and security risks that extend far beyond simple cost-benefit calculations. The debate mirrors previous controversies over Chinese electric vehicles, where cheap imports raised concerns about subsidized exports and long-term industrial hollowing.

The Illusion of Self-Hosted Security

For developers and CTOs, the primary appeal of Chinese open-weight models is the ability to run them on private servers, theoretically isolating data from foreign vendors. However, Ross Burley, director of the Centre for Information Resilience (CIR), warns that this technical isolation is a false sense of security. He notes that while running a model locally prevents external API access, it does not remove embedded political censorship or repair inherent weaknesses in security defenses. The software’s architecture often retains the ideological constraints of its origin, regardless of where it is deployed.

NIST Findings and Data Sovereignty

Concrete evidence supports these concerns, with an evaluation report by the US National Institute of Standards and Technology (NIST) finding political censorship in downloaded DeepSeek models even when running on independent infrastructure. This means that hosting these models in the UK does not automatically sanitize their output or align them with local regulatory standards. Furthermore, DeepSeek’s policy explicitly states that it collects prompts and uploaded material, processes personal data in China, and permits disclosure to Chinese authorities, creating a direct pipeline from local servers to Beijing.

Strategic Implications for Public Services

Professor Michael Wooldridge of the University of Oxford emphasizes that it is not unreasonable to assume data inputted into these chatbots could be shared with the Chinese state, a risk that is unacceptable for sensitive government operations. The broader argument is that relying on Chinese technology, from EVs to AI, undermines the security of domestic data and hollows out local industry. Instead of accepting the low-cost option, the report suggests the UK should focus on taming US AI providers while actively supporting domestic and European suppliers to build a resilient, secure tech stack.

Key Takeaways

  • Chinese open-weight models (DeepSeek, Xiaomi) offer no ongoing license fees but embed political censorship that persists even when self-hosted.
  • NIST evaluations confirm that downloading models does not remove ideological constraints or security weaknesses inherent in the software.
  • DeepSeek’s data policy allows for the collection and processing of user prompts and personal data within China, with disclosure to local authorities.
  • Experts argue that cheap Chinese tech risks hollowing out UK industry and compromising data sovereignty for public services.

The Bottom Line

Self-hosting Chinese AI models is a security theater that fails to mitigate the inherent risks of political censorship and data extraction. Britain must prioritize data sovereignty and industrial resilience over the deceptive allure of low-cost infrastructure.