The UK financial regulator will set out its response to the latest advancements in frontier artificial intelligence models.
The UK financial regulator is planning to use artificial intelligence (AI) to automate most of its supervisory and enforcement activities, transforming itself into an "agentic regulator," though it currently has no plans to impose new regulatory rules on the financial industry.
The Financial Conduct Authority (FCA) will announce its response to the latest frontier AI models next week. At that time, the agency's Chief Data, Information and Intelligence Officer, Jessica Rusu, will deliver a speech outlining the relevant plans.
Frontier AI models have become one of the most closely watched topics among financial services executives and regulators. All parties are working to address the potential transformation this technology could bring to the industry, as well as the new risks that accompany it.
Rusu is expected to explain how the FCA is responding to recommendations put forward by former senior regulatory official Sheldon Mills. In July of this year, Mills called on the FCA to take action to regulate the use of AI in the financial sector. The assessment report chaired by Mills pointed out that the FCA needs broader powers to keep pace with the development of AI. The report urged UK authorities to review whether the use of ChatGPT, Claude, Gemini and other large language models should be brought within the scope of regulation.
Wealth management firms are concerned that millions of people have already begun using AI to obtain personal financial advice, yet these models remain unregulated, and if problems arise, users have no avenue for compensation.
A study commissioned by Mills found that one in five UK adults is already willing to let AI models make financial decisions on their behalf, such as decisions about saving or borrowing.
FCA Chief Executive Nikhil Rathi has repeatedly expressed skepticism about whether new rules specifically targeting AI in finance are necessary. This week, Rathi told lawmakers at a hearing held by the UK Parliament's Treasury Committee that although his views on AI "have been evolving," he still does not believe regulators need to "acquire substantial additional powers unless we have genuinely demonstrated that we have fully utilized our existing powers."
Rathi said he is often asked whether the FCA should regulate AI models. He said: "At present, if a model merely provides information without commercially offering personalized advice that facilitates transactions, then whether in financial advice, insurance or other markets, it does not entirely fall within the existing regulatory scope."
"But I think this will become a very important topic of discussion: where exactly should we draw the regulatory boundary?"
Another recommendation in the assessment report is that the FCA should "build and adopt an AI-driven agentic regulatory model," so that it can still conduct effective oversight as an increasing number of financial services activities are carried out by autonomous agents with diminishing human involvement.
Rusu said at the Treasury Committee hearing that AI is particularly effective in analyzing and summarizing "unstructured text" that firms provide to regulators through emails, documents and interviews.
She said: "Becoming an agentic regulator is part of our overall goal of achieving smarter regulation."
"When considering the various ways to use this technology for intelligence gathering and case handling optimization, we have already begun exploring how to embed it into the agency's authorization approvals, day-to-day supervision and enforcement work."
The FCA is deploying AI while ensuring that supervisors can oversee and guide the relevant activities. However, Rusu said the agency wants to "go further, automating in all areas where feasible… and conducting regulatory interventions at scale."
Rusu said the FCA has been conducting AI field testing, visiting financial services firms to understand how they deploy and apply the technology. She added that the regulator has already "developed a deep understanding of what works well and what poses problems," and will publish the relevant findings next week.
Comments