FCA AI review puts regulatory data in focus for financial firms

The UK Financial Conduct Authority’s Mills Review examined how artificial intelligence could reshape retail financial services, including fraud controls, advice, customer interactions and supervision. Rohini Gupta, chief executive of FinregE, said in a commentary published on Finextra that the review points to regulatory intelligence as a potential competitive advantage as AI becomes more embedded in financial firms.

The review, as described by Gupta, considers a market in which AI plays a larger role across the customer journey. That includes helping consumers compare financial products, receive more tailored guidance and, over time, delegate more decisions to intelligent agents.

Gupta said the FCA also considers how AI could help address the advice gap, support broader access to financial services and allow firms to respond more closely to customer needs. Those possible benefits, she said, sit alongside significant responsibilities for regulated institutions.

Risks identified in the review

According to Gupta, the FCA review highlights concerns already familiar to boards, risk teams and supervisors: governance, transparency, explainability, cyber resilience, operational resilience and consumer protection.

It also addresses risks that could grow as AI tools become more capable. Gupta cited concentration risk where firms depend on a limited group of AI model providers, cloud companies and technology platforms. She also pointed to more sophisticated fraud, including deepfakes, and the difficulty of assigning accountability when more autonomous systems influence or execute financial decisions.

Gupta argued that these issues indicate a change in the operating model for compliance. Historically, she said, firms have handled regulation through largely human-led processes: rules are issued, legal and compliance teams interpret them, internal policies are revised, controls are adjusted and evidence is later produced for supervisors.

AI changes the tempo of that process, in Gupta’s analysis. Systems that support or automate decisions can operate continuously, while regulation and supervisory expectations keep developing. She said compliance functions built around static documents, spreadsheets and fragmented processes may struggle where both firms and regulators use more data-driven tools.

Machine-readable obligations

FinregE’s position, as set out by Gupta, is that financial firms will need regulatory intelligence that can be used by both employees and machines. In practical terms, that means translating legal and regulatory obligations into structured, current and authoritative information that can inform AI-enabled decisions.

Gupta said AI used in areas such as affordability checks, customer communications, financial crime controls, product launches and Consumer Duty processes must be connected to an accurate regulatory context. Without that link, she argued, a model could generate outputs that appear technically sophisticated while failing to meet regulatory requirements.

The issue becomes sharper with agentic AI, according to Gupta. As tools move from assisting staff to taking actions on behalf of firms or customers, governance must cover more than the workings of the model itself. Firms also need to understand the regulatory basis for recommendations, automated controls, customer communications and delegated decisions.

Gupta also noted that the FCA review recognises regulators’ own interest in using AI to strengthen supervision, improve analysis and detect emerging risks. In her view, that could raise expectations for firms to modernise their own regulatory operations.

The review’s central policy concern, as presented by Gupta, is maintaining trust and consumer protection while allowing useful AI adoption. FinregE’s conclusion is that firms investing in reliable, continuously updated regulatory intelligence may be better placed to show supervisors how AI-enabled activity remains within approved boundaries.

This story draws on original reporting from Finextra Research.