David Lammy’s department used an AI model created by tech giant Alibaba to analyse criminal records, in what was the first-known use of a Chinese AI system within Whitehall.

The revelation risks prompting fresh tensions with Washington after the Trump administration last week blacklisted Alibaba because of national security concerns.

Last week, the US department of defence added the tech business to a list of companies with alleged ties to the Chinese military – a claim the company denies.

The MoJ, run by the Deputy Prime Minister, used Alibaba’s Qwen AI model for a study of whether addiction treatment services prevented criminals from reoffending. The work involved assessing tens of thousands of criminals’ files, including the type of conviction and nature of treatment.

Qwen is an open-source model, meaning users can copy the code and run it on their own computer servers to prevent data from being sent to the developer. The MoJ said the data were anonymised and that it had run the system on its own “analytical platform”, meaning data would not have been sent to Alibaba servers.

However, wider concerns have been raised about the use of Chinese AI. The US government has warned against using such systems, which experts have said are biased towards Beijing’s viewpoints and could make the West more dependent on China.

‘This will come back to bite us’

Luke de Pulford, the executive director of the Inter-Parliamentary Alliance on China said: “Using Qwen opens the door to UK reliance on AI systems controlled by foreign adversaries, with attendant security vulnerabilities.”

He added: “Depending on Chinese AI to run key parts of the state is an unnecessary risk that will come back to bite us.

“Playing into Beijing’s AI propaganda war and trusting their software is an unforced error that does nothing to advance national security.”

A review from the US-based China Media Project found that Qwen was trained “to subtly deliver positive messages about the country to an international audience”.

US politicians have said that using Chinese models risks giving Beijing the upper hand in a global battle to set the rules of AI and have pressured companies using Qwen to drop the system.

A report from the US-China Economic and Security Review Commission, a branch of the US Congress, warned this year that China “dominates the global open [source] ecosystem” through models such as Qwen and DeepSeek. This was helping the country to develop a “self-reinforcing competitive advantage”, the report said.

Beijing’s global adoption strategy

Chinese tech companies, with support from Beijing, have pursued a strategy of making their systems free to download and use, as well as significantly cheaper to run, in an attempt to encourage their use around the world.

The cheaper systems have become increasingly popular as broad adoption of AI has pushed up its cost.

The MoJ’s use of Qwen was disclosed within technical notes on a study into whether addiction treatment prevented reoffending, published in March. It said the department had also considered using Anthropic’s Claude model but opted for Qwen.

The department has been an enthusiastic adopter of AI, striking a deal with OpenAI to have data stored on British soil and exploring how the technology can be used to provide court transcripts and help judges write decisions.

The Government’s AI guidelines do not ban the use of Chinese models, although the DeepSeek app is banned on government phones.

Alibaba was placed last week on the Pentagon’s “Chinese military companies” list. Alibaba said there was no basis for its inclusion. The MoJ study was carried out before the designation.

Peter Kyle, the Business Secretary, has said it is imperative that the AI race be won by “Western, liberal, democratic countries”.

“Government does have agency in how this technology is developed and deployed and consumed,” he said last year. “We need to use that agency to reinforce our democratic principles, our liberal values and our democratic way of life.”

The Ministry of Justice said no data entered into the system were sent to China or Alibaba.

“Protecting sensitive information is a fundamental requirement of government-backed research reports, and all data use is subject to strict legal, security and governance controls,” a spokesman said.

“This AI model is being used locally in a secure UK-based cloud environment, with no data transmitted outside of UK servers.”

Source link