UK Banks: New Testing Regime for AI Systems
The UK Government is considering the introduction of a standardized testing procedure for general-purpose Artificial Intelligence (AI) models used by UK lenders. The move comes in response to concerns raised by the Bank of England (BoE) last year over the evaluation methods of such AI models.
Harriet Rees: Proposing a Common AI Testing Framework
Harriet Rees, chief information officer at Starling Bank, presented the idea to the Department for Science, Innovation and Technology. Currently serving as the government’s financial services AI “champion” and co-chair of the BoE’s AI task force, Rees expressed the need for a common, independent assessment of AI models.
“Lots of firms are using [AI models] and we can assume that [they] have done the necessary due diligence and, therefore, hopefully we’re happy. But we’ve not done that independent assessment,” she stated.
Addressing Inconsistencies and Ensuring Standards
The proposed testing regime aims to eliminate redundant work across firms, bring consistency to the testing process, and ensure that AI algorithms developed in the US meet the necessary standard. The need for a common assessment was underscored during two AI meetings held by the BoE’s Prudential Regulation Authority in October, when it was pointed out that AI model monitoring was “not frequent enough”.
Rees added: “Given our reliance on US models, it would give [the government] the comfort that they’ve at least looked at [the models] and they know that they all are at a certain standard.”
Current State of AI Model Assessments
At present, there is no legal obligation for AI systems to undergo assessment before being deployed in regulated sectors, although banks do carry out their own reviews. Several groups, including OpenAI and Anthropic, have voluntarily submitted their models, such as ChatGPT and Claude, to the AI Security Institute (AISI), a government unit focused on testing advanced AI systems and researching related risks.
Choosing the Right Regulator
Rees argued that the responsibility for examining these general-purpose AI models should not fall on a single sector regulator, as their use extends beyond financial services. She suggested that AISI was the “most obvious body” for this role. However, a government spokesperson indicated that AISI, which is focused on frontier AI security research, was unlikely to be given this task.
Independent Oversight: A Fail-Safe Measure
Rees, who has previously voiced concerns about corporate dependence on US technology, emphasized that oversight by an independent body would not replace lenders’ own checks. Instead, it would serve as a “fail-safe” and offer reassurance about the quality and reliability of the AI models. The BoE declined to comment on this issue.
Overall, the introduction of a common testing regime for AI models used by UK lenders could provide a more standardized, independent evaluation process that ensures the reliability and quality of these AI systems.
For more details on this development, visit the source article here.
