Out-Law News 3 min. read
13 May 2025, 8:21 am
Financial services firms will be able to live-test artificial intelligence (AI) models with the help of UK regulators, under an initiative described as “pioneering and ambitious” by experts in fintech and financial regulation.
Luke Scanlon and Daniela Ivanova of Pinsent Masons were commenting after the Financial Conduct Authority (FCA) set out proposals (17-page / 395KB PDF) to develop a new AI live testing service through its existing AI Lab. The proposals are open to consultation until 10 June 2025 with plans for the new service to start operating from September this year.
For firms, the new service is aimed at supporting them with implementing consumer-facing AI tools. Access to the new service will be subject to eligibility criteria, with the process “designed to maximise value for firms looking to deploy established proofs of concept”, the FCA said, adding that it will not be suitable for “firms just embarking upon an early proofs of concept phase, or firms making operational improvements that don’t require regulatory support”.
The FCA said: “We will provide participants with access to appropriate AI expertise in addition to regulatory expertise and seek to understand the nature of any emerging regulatory challenges they may face. AI Live Testing is designed to be exploratory in nature and does not seek to cover AI auditing, certification, regulatory compliance with other frameworks or corporate governance questions.”
“Participant firms need to demonstrate that they have carried out effective pre-deployment testing and are ready to deploy into controlled live market environments, interacting with consumers. We will work with firms to make sure they have appropriate risk identification and safeguards in place, including an approach to addressing any potential harms,” it said.
“As part of AI testing, the FCA has a range of tools it can use to provide firms with regulatory comfort, where appropriate to do so. Similar to the regulatory sandbox, this may include providing individual guidance, waivers or modifications to requirements. Where market or sector wide issues are identified, the insights gathered may help inform our regulatory approach more generally, including whether any further clarifications or guidance might be helpful,” the regulator added.
While the testing undertaken will be specific to the AI models individual firms seek to implement, the FCA said it plans to share non-confidential findings from its work with the market. The regulator set out a list of questions it would like to explore in this regard, which include: how input-output validation might offer comfort that gen-AI output meets regulatory expectations; how to assess if an AI model is sufficiently robust; how an AI model’s bias can be measured and addressed; and how poor or unintended AI model outcomes will be addressed when they arise.
Scanlon said: “AI live testing is pioneering and ambitious – the FCA has promised to provide participating firms with appropriate AI expertise in addition to regulatory expertise.”
Scanlon said that the illustrative example accompanying the FCA’s engagement paper is attractive as it involves AI being used for loss provisioning and credit provisioning by lenders, which impacts directly on financial resources requirements, though not an external consumer-facing application per se.
“It will be exciting to see what consumer-facing applications are sanctioned by the FCA via AI Live Testing and how much information the FCA will be able to provide about their workings in order to assist other firms, which are in reality potential competitors,” Scanlon said.
Ivanova added: “Firms may be forgiven for being hesitant to adopt AI for external consumer-facing applications, given an outcomes-based regulatory approach. Their experience with outcomes-based regulation so far centres on the Consumer Duty, which has seen a steady cadence of FCA thematic reviews, updates on good and poor practice, and follow-up actions to self-assess and address poor practice and implement good practice. However, the FCA is offering an olive branch. It intends to use insights gathered in AI Live Testing to publish best practice and guidance for the rest of the industry.”
The FCA intends for the new AI live testing service to run for 12 months initially, before it is evaluated. It said the initiative supports the UK government’s growth agenda – and its AI opportunities plan, specifically – and aligns with moves that other regulators are making around the world, including in the US and Singapore, to support businesses with AI adoption.
The proposed service will be distinct from other pro-innovation initiatives the FCA operates – its ‘sandbox’ schemes already support firms with trialling new technologies or processes.