UK weighs standard tests for AI models used by banks

In short: The UK government is considering standard, independent testing for general AI models that banks use.

What happened

The British government is weighing a plan to run standard tests on “general purpose” AI models used by UK lenders. General purpose AI models are systems trained to handle many tasks, like writing text or sorting information (similar to a very flexible calculator that can do lots of different jobs).

The idea was proposed to the Department for Science, Innovation and Technology by Harriet Rees, chief information officer at Starling Bank. Rees said many firms assume the AI models they use have been checked properly, but there is no independent assessment.

The proposal follows warnings from the Bank of England’s Prudential Regulation Authority, which regulates lenders. At meetings in October, banks were told their AI model monitoring was “not frequent enough,” according to slides from those events.

Rees said independent testing could reduce duplicated work and inconsistent checks across the industry. She also said it could help address the UK banking sector’s reliance on US-made AI models by confirming they meet a minimum standard.

There is currently no law requiring AI models to be evaluated before being used in regulated industries, although banks do their own testing. Some AI companies, including OpenAI and Anthropic, have voluntarily provided models for evaluation by the UK’s AI Security Institute.

A UK government spokesperson said the AI Security Institute is focused on security research and that the government is not exploring expanding it into testing third-party AI models.

Why it matters

Banks use AI in areas that can affect people directly, like spotting fraud or deciding whether an application looks risky. Standard tests would be like an MOT for shared AI tools, meant to provide a basic safety check before many banks rely on the same systems.

Source: Financial Times