Starling Bank’s chief information officer Harriet Rees has asked the Department for Science, Innovation and Technology to introduce a standardized test for general‑purpose AI models used by every UK lender. The proposal follows a Bank of England warning last year that banks’ own AI evaluations were insufficient. A uniform regime would curb duplicated effort and ensure imported US models meet a baseline standard across the financial sector.

Rees, also a government AI champion and co‑chair of the BoE’s AI task force, argues that testing should not sit with a single regulator because the models span multiple sectors. She points to the AI Security Institute, the government’s frontier‑AI research unit, which already receives voluntary model submissions from firms such as OpenAI and Anthropic.

The government has so far signaled that the Institute will not take on formal assurance duties, keeping its focus on security research. Rees says an independent assessment would act as a fail‑safe, complementing banks’ own checks rather than replacing them. With no legal requirement for pre‑deployment AI testing, the move could become the de‑facto benchmark for UK lenders.