FCA plans to embrace AI to automate financial regulation

Britain’s financial watchdog is planning to become an “agentic regulator” by harnessing AI to automate much of its supervisory and enforcement activities, but is stopping short of imposing new rules on the sector.

The Financial Conduct Authority will next week unveil its response to the latest frontier AI models in a speech by Jessica Rusu, chief data, information and intelligence officer.

It comes as frontier AI models have risen to the top of the agenda for financial services executives and regulators, who are grappling with the technology’s potential to transform the sector and the new risks it brings.

Rusu is expected to explain how the FCA will respond to the recommendations of Sheldon Mills, a former senior executive at the regulator, who in July called on it to take action on the use of AI in finance.

The Mills Review said the watchdog would need greater powers to stay on top of AI and urged UK authorities to review whether the use of ChatGPT, Claude, Gemini and other large language models should be subject to their rules.

Wealth managers worry that millions of people are already turning to AI for personal finance advice even though the models are not covered by regulation and there is no recourse to compensation if things go wrong.

Research commissioned by Mills found a fifth of UK adults were already open to using AI models to make financial decisions for them, such as on savings or borrowing.

Nikhil Rathi, FCA chief executive, has repeatedly expressed his scepticism about the need for new rules on AI in finance. He told MPs at a Treasury select committee hearing this week that while his thinking on AI “has been evolving”, he was still not convinced the regulator needed “lots more powers until we have really demonstrated that we have used our powers to the fullest extent”.

Rathi said he was often asked whether the FCA should regulate AI models.

“At the moment, where a model is providing information but is not providing an individualised recommendation which is leading to a transaction by way of business, it does not fall fully within the regulated perimeter, whether that is for financial advice, insurance or other markets,” he said. “But I think that is going to be a very live discussion: where do you draw that boundary?”

Another of the review’s recommendations was for the watchdog to “build and adopt an AI-enabled agentic supervisory model” to stay on top of financial services activities as they are increasingly carried out by autonomous agents with less input from humans.

Rusu told MPs at the Treasury committee that AI was particularly effective in analysing and summarising the “unstructured text” that companies provide to the regulator in emails, documents and interviews.

“Becoming an agentic regulator is part of our overall ambition of being a smarter regulator,” she said. “As we consider the various ways we can use that for intelligence gathering and case optimisation, we have been able to look at ways to embed that in authorisations, supervision and enforcement work.”

The regulator is deploying AI in ways that allow its officials to oversee and direct its activities, but Rusu said it aimed to go “further than that, to automate wherever we can . . . and to intervene at scale”.

Rusu said the regulator had been carrying out AI live testing by visiting financial services firms to examine how they were implementing the technology, adding that it had “learnt a lot about good and poor practices” and would publish its findings next week.