Sixty-two percent of financial services firms have already deployed AI agents, and 93 percent of those firms let the agents act with some degree of autonomy. Those figures, from a June report by the Cloud Security Alliance, describe an industry that has quietly crossed a line. Software is no longer just recommending decisions inside banks. It is making them. Now the regulators are catching up, and the United Kingdom’s Financial Conduct Authority has just published the clearest statement yet of what that will mean.
What the Mills Review actually says
The FCA’s landmark review of AI in retail financial services, led by Sheldon Mills, the regulator’s executive director of consumers and competition, argues that the industry is shifting from human-led to AI-enabled, and that regulation will play a decisive role in either speeding or restricting that shift. The review identifies four ways the technology will remake the sector: growing autonomy and centrality inside business operations, customer journeys that become agent-led, new dynamics in market competition, and a rising threat from AI-enabled fraud.
The most striking argument concerns where risk lives. “As autonomy grows, the nature of regulatory risk changes,” Mills wrote in the review’s foreword. His point is that as firms and consumers delegate more decisions to AI systems, the danger stops being contained within a single institution and starts spreading across the financial system itself. A misfiring model at one bank is a problem. Thousands of autonomous agents interacting across dozens of firms is a different category of problem entirely.
Banks are not waiting for permission
The review lands in the middle of an arms race. TD Bank’s U.S. arm is assessing how AI agents can rework its business processes. BNY has deployed hundreds of digital employees into its workflows. In Australia, Commonwealth Bank hired Ranil Boteju, previously chief data and analytics officer at Lloyds Banking Group, as its first chief AI officer earlier this year, and Lloyds responded by handing Sameer Gupta a newly created chief data and AI officer role to lead its own strategy.
Executive appointments like these are the tell. When banks start creating C-suite roles for a technology, it has stopped being an experiment and become infrastructure.
How the regulator plans to respond
The FCA’s answer is not to slow the technology down but to change how it supervises. The review commits the regulator to adapting its framework for AI-enabled financial systems, strengthening oversight at the system-wide level rather than firm by firm, and building what it calls an agentic supervisory model, which would use AI to supervise AI. For a body often criticized as slow-moving, that is an unusually forward-leaning position.
The review also carries a warning for the firms themselves. Governance, it argues, should be treated as a core enabler of AI capability rather than a compliance chore. Institutions will need to manage their growing dependence on third parties, including the companies that build the underlying models. Some firms may even allow customer-owned agents or external systems to interact directly with their infrastructure, which raises hard questions about who is allowed to access data, trigger workflows and initiate actions inside a bank that no longer has a human at every gate.
The delegation problem
Striking a balance between enabling delegation and managing autonomy is, by the FCA’s own account, the central challenge it now faces. That framing matters beyond Britain. The FCA has historically set the tone for financial regulators worldwide, and supervisors in the United States have already begun asking banks detailed questions about how they use AI in lending, know-your-customer checks and sanctions screening.
The industry spent the past two years proving that AI agents work in finance. The next two will be spent deciding who is accountable when they do not. Watch for the FCA’s framework changes to start surfacing in consultation papers over the coming months, because whatever rules emerge in London will not stay there. For more coverage of AI in finance, visit Mylistingo.
