A high-ranking official from the FCA suggests that the UK ought to consider directly regulating AI models.

      Britain should contemplate the possibility of regulating large language models like ChatGPT, Claude, and Gemini, as they increasingly affect consumers' financial decision-making, according to a senior official at the Financial Conduct Authority (FCA). Sheldon Mills, an executive director at the FCA, emphasized that the current regulations will need to adapt as companies increasingly rely on a limited number of tech providers, a situation he cautioned could lead to systemic risk. This is a significant pronouncement from a regulator that has largely adhered to the UK's pro-innovation, principles-based approach up until now.

      Mills highlighted a specific concern: over a quarter of UK consumers already rely on platforms such as OpenAI's ChatGPT, Anthropic’s Claude, and Google’s Gemini for financial guidance, often unaware that the safeguards provided by regulated financial services do not apply to these technologies. Consumers are making financial decisions based on software that operates outside the jurisdiction of the FCA.

      This disparity is at the heart of the issue. When a regulated advisor provides subpar advice, consumers have avenues for recourse; however, when a general-purpose chatbot does the same, accountability is much less clear.

      Mills contended that technology has quietly assumed a role that the regulatory framework never predicted, and ignoring this reality presents its own risks. His suggested solution is procedural yet direct. Mills advised that the FCA should determine, within the next three to six months, whether to "secure and adapt" the regulatory boundary by assessing the scope, characteristics, and effects of the general-purpose models that currently lie beyond it. This does not immediately call for new regulations but sets a timeline for evaluating whether foundational models should be incorporated into financial regulation.

      He elaborated on potential oversight measures, proposing new powers for the FCA that would require firms to clarify how their AI models make decisions, audit algorithms for fairness, and impose penalties on systems that inflict harm on consumers. Each of these suggestions addresses a significant challenge in AI governance, as many models tend to be opaque even to the companies using them.

      Mills’s remarks conflict somewhat with the government’s broader approach. The UK has intentionally steered clear of establishing specific AI legislation, opting instead to assign oversight to existing regulators to maintain a pro-innovation advantage over the EU.

      While Mills is not advocating for a UK AI Act, he believes there is a significant gap in the sector-by-sector regulatory model concerning general-purpose systems, indicating that the FCA may need to address it.

      There is an ongoing debate regarding where responsibility should lie, which remains unresolved. Regulating the models themselves instead of just the firms that utilize them would represent a significant shift in strategy, conflicting with the UK’s tendency to favor a lighter regulatory touch. Additionally, it faces the challenge that the largest models are primarily developed by a few American companies that the FCA does not regulate.

      Mills's point about concentration is arguably the more critical issue. If most regulated firms become reliant on a few model providers, any failure in one of those systems could have widespread effects across the financial sector, a type of correlated risk that regulators typically seek to eliminate. This concern is well-known in cloud computing and is now extending to the models developed upon it.

      At this juncture, Mills’s statements serve as a signal rather than a formal policy. However, they come at a time when governments globally are acknowledging that AI is advancing faster than the regulations intended to govern it, and a financial regulator suggesting that the regulatory boundary may need adjustment is significant. The coming months will reveal whether the FCA perceives this as a genuine review or merely as rhetoric.