MARKET INSIGHT · EARNINGS INTELLIGENCEPolaris Earnings CalendarTrack market technicals, follow the S&P 500 outlook, stay ahead of company earnings, and explore crypto technicals—all in one place.

POLARIS EARNINGS CALENDAR

Britain weighs the security cost of China’s fast-spreading open-weight AI

Britain weighs the security cost of China’s fast-spreading open-weight AI

Low-cost Chinese artificial-intelligence models are expanding their reach because businesses can download and adapt their core software without paying the continuing licence fees charged by many proprietary systems. DeepSeek’s models, along with products from other Chinese developers, offer British organisations a cheaper alternative to closed US systems.

That price advantage comes with documented concerns about censorship and security.

The evaluation also found political restrictions embedded in downloaded DeepSeek models, rather than only in the company’s online service. When tested in English and Chinese, the models were more likely than the US reference systems to reproduce inaccurate or misleading narratives aligned with the Chinese Communist Party. The findings mean that running the software on British servers would not, on its own, eliminate censorship.

DeepSeek’s own privacy policy says the service can collect prompts, uploaded files, feedback and chat histories. It identifies the Chinese company as the data controller and says information may be shared with law-enforcement agencies or public authorities when required by law or government requests.

The policy creates a separate risk for organisations that put sensitive commercial, personal or public-sector information into the online chatbot. Local hosting can reduce exposure to the provider’s servers, but it cannot automatically remove restrictions built into model behaviour or correct weaknesses in the software’s security.

The technology is nevertheless spreading.

DeepSeek models were competitive in some science, knowledge and mathematics tests, but US models had wider advantages in cyber and software-engineering tasks.

For Britain, the choice is therefore not simply between free software and expensive software. It also concerns which political restrictions, data practices and security weaknesses may be carried into systems used by companies and public services. The available evidence supports tighter scrutiny of Chinese AI while Britain develops domestic and European alternatives and applies stronger safeguards to AI from all suppliers.

Read the full account of the story

More from the AI desk

More from AI · Back to Polaris Earnings Calendar