Britain's financial regulator is urging a prompt reassessment of whether widely available large language models should be treated as general-purpose artificial intelligence that sits within regulatory oversight, a senior official said.
The Financial Conduct Authority's Executive Director, Sheldon Mills, told industry stakeholders that tools such as OpenAI's ChatGPT, Anthropic's Claude and Google's Gemini are playing an expanding role in consumer financial decision-making. Mills said that, given that trend, the FCA should examine the scale, nature and impact of those models - many of which currently operate outside the regulator's perimeter - and consider whether the perimeter should be "secured and adapted" within the next three to six months.
Mills' evaluation is part of a broader FCA review into how artificial intelligence affects the financial sector. The review notes a spectrum of risks regulators are confronting worldwide, from cyber and operational vulnerabilities linked to frontier AI models such as Anthropic's Mythos to the difficulties associated with agentic systems that can act with limited human oversight.
Highlighting consumer behaviour, the FCA's findings show that more than a quarter of UK consumers place trust in large language models for financial advice, despite a limited awareness that the consumer protections that apply to regulated financial services do not extend to interactions with these unregulated tools. The report identified widespread consumer use alongside a gap in understanding about regulatory safeguards.
The review also examined adoption trends within financial firms. A recent survey cited in the review found that 81% of financial firms globally are using AI at some level, and roughly 40% are at more advanced stages of scaling or transformation. While many of the current applications are concentrated in lower-risk, back-office functions, the FCA observed British firms increasingly deploying AI in roles that interface directly with customers, including complaint handling and investment guidance.
Mills warned that rapid, widespread adoption could leave firms dependent on a small number of technology providers for essential operational capabilities. Such common reliance - whether on the same large language models, cloud services or other technology infrastructure - could produce correlated behaviour, herding dynamics and common points of failure across the financial system, the review said.
Responding to the review's findings, Ashley Alder, Chair of the FCA, said: "We need to keep pace with a rapidly changing environment and the principles-based, outcomes focussed approach weve taken on AI".
Context for industry participants
- Financial services firms should assess dependency risks tied to small numbers of AI and cloud providers, particularly as customer-facing use cases expand.
- Consumer-facing deployments increase the potential for misaligned expectations about regulatory protection when users seek financial advice from models outside the regulatory perimeter.
- The FCA is contemplating perimeter adjustments within a three- to six-month window to address these developments.