A high-ranking official from the FCA states that the UK should consider the direct regulation of AI models.

      Britain should examine whether large language models like ChatGPT, Claude, and Gemini require regulation as they become increasingly influential in consumer financial decisions, according to a senior official from the Financial Conduct Authority (FCA).

      Sheldon Mills, an executive director at the FCA, stated that the current regulatory framework will need to adapt as companies rely on a limited number of technology providers, which he cautioned could pose a systemic risk. This marks a significant stance from a regulator that has primarily adhered to the UK's pro-innovation, principles-based approach.

      Mills highlighted a specific concern: over a quarter of UK consumers already trust tools like OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini for financial advice, often without realizing that the protections associated with regulated financial services do not apply to these tools. Individuals are making financial decisions based on software that operates entirely outside the FCA's regulatory scope.

      This disconnect is central to his argument. When a regulated advisor provides poor advice, consumers have options for recourse; however, the accountability is much less clear when a general-purpose chatbot does the same.

      Mills emphasized that the technology has taken on a role that the existing regulatory framework did not foresee, and ignoring this reality poses its own risks.

      He suggested a procedural yet specific solution. Mills recommended that the FCA determine within the next three to six months whether to "secure and adapt" the regulatory boundaries by examining the scale, nature, and impact of the general-purpose models that currently lie outside it. While this doesn't call for immediate regulations, it puts a timeline on the discussion regarding the inclusion of foundational models in financial regulation.

      He further outlined what oversight might eventually entail, proposing new powers for the FCA to require firms to clarify how their AI models make decisions, to conduct fairness audits of algorithms, and to impose penalties on systems that harm consumers.

      Each of these suggestions addresses a significant challenge in AI governance, as many models remain opaque even to the companies that utilize them.

      This intervention contrasts somewhat with the government's overall approach. The UK has intentionally avoided creating a specific AI law, opting instead to assign oversight to existing regulators, aiming to maintain a pro-innovation edge over the EU.

      Mills is not advocating for a UK AI Act but indicates that the sector-by-sector model has a gap concerning general-purpose systems, suggesting that the FCA might need to address this.

      A current debate exists regarding where accountability should lie, and it remains unresolved. Regulating the models themselves, rather than just the firms that use them, would represent a significant shift in approach, challenging the UK's established light-touch regulatory inclination. Additionally, this faces the practical issue that the largest models are developed by a few American companies that the FCA does not oversee.

      The issue of concentration that Mills raised is arguably more pressing. If the majority of regulated firms rely on the same two or three model providers, a failure or defect in one of those systems could have widespread repercussions across the financial sector, generating the kind of correlated risks regulators typically seek to eliminate. This concern is familiar in cloud computing and is now extending to the models built on that foundation.

      At present, Mills's remarks serve as a signal rather than a specific policy. However, they arrive at a time when governments worldwide acknowledge that AI is outpacing the regulations intended to govern it, and a financial regulator indicating that adjustments may be necessary is noteworthy. The coming months will reveal whether the FCA treats this as a serious review or merely a discussion point.