From the cars we drive to the films we watch, AI is impacting all areas of our lives. But some of those areas are heavily regulated, and the consequences could be profound. Nowhere is that more true than financial advice and AI. Although 91% of people don’t access financial advice from people, 100% have access to LLMs like ChatGPT and Claude, and they are asking AI to help with financial decisions. And AI is offering financial advice – one LLM just told me to get a five year fixed rate mortgage with HSBC.
Sheldon Mills, executive director of the FCA, is conducting a review of AI and retail financial services. He recently spoke to AI providers, financial advisers, wealth managers and law firms at a Prysm Global roundtable, and the exchange of views was fascinating.
The use of AI to answer questions in real time is commonplace; we also see AI being used to automate many tasks and processes. There is now clearly a move to use AI in client-facing roles. Data-led firms are moving quite quickly to agentic AI: the technical capability is already there for an AI agent to move bank accounts on your behalf.
There are huge opportunities here for consumer benefit – for example, more people getting better advice, cheaper. Advice that might now cost £2000 may be given for £20. It is easier for an AI agent than a person to scan the entire market quickly. Consumer experience and usability could improve radically. Compliance with rules could be guaranteed.
But there are also huge risks. The AI agents are not regulated by the FCA, and so nor is their advice. The consumer protection guardrails risk being ripped off. Personal use of AI seems to be outpacing institutional comfort levels, suggesting that consumer expectations of AI could shift faster that firms’ and regulators’ risk appetites.
What are the limits to this? Where in the future will consumers get their advice from – regulated people or unregulated computer programmes? Gaps are clearly opening up. It is not in consumer interest if regulation inadvertently pushes them to unregulated advice. The FCA has defined its regulation of advice in terms of traditional legal entities and definitions and standards for the provision of advice which have contributed to the emergence of the huge advice gap. The FCA is now changing its definitions of advice to make it less onerous and risky to provide advice.
Meanwhile, Open AI have said that over 200 million people every month ask ChatGPT questions on investments, budgeting and financial planning. Frontier AI providers and platform providers are bringing their own solutions. In the US, OpenAI has just launched a financial planning tool that lets users connect their financial accounts and ask questions based on their financial context. None of the answers they get from AI in the UK are regulated.
Underlying the speed of deployment and take-up and the role of regulation are questions of trust. How quickly will consumers decide to trust AI with their finances? What can AI firms do to build trust? Everyone needs to remember the adage that trust arrives on foot but leaves on horseback at a gallop. What is the role of the regulator in ensuring that the AI is trustworthy without stifling innovation and the benefits.
A key question then is where should the FCA regulatory perimeter lie? Do we need to move from regulating traditional legal entities (which is easier) to regulating models (which may make more sense, but would be more difficult to enforce). Should AI agents be regulated? There is a senior managers regime for people, but should there be a senior managers regime for AI agents? The current view is “no”, but will that still be the case in 5 years’ time when there may be an autonomous AI economy?
There are risks of harm as well as benefits from AI in advice and financial services.
There are challenges of preventing answers and actions based on poor or even poisoned data. Data protection is an acute issue legally – including whether big data providers are doing enough to address the deliberately false poisoning of personal data and how personal data is protected from harmful usage. Open Finance – the authorised sharing of personal data – is going much too slowly for many in the industry. AI could open up whole new opportunities for Open Finance, helping to reach its so far unfulfilled promise.
There are issues of bias and explainability. There are potential access barriers caused by hyper-personalisation which could undermine the pooling of risk that is the basis of the insurance market.
In the increasingly problematic area of fraud, AI brings benefits and risks. Fraudsters can use AI to target victims more convincingly. But investigators can use AI to do the investigations humans do but in a fraction of the time. In future it might be possible to review all payments for fraud in real time, not just when there is an alert flagging suspicious activity.
This is a whole new frontier of regulation in financial services. It will be important to get the right balance – to ensure consumers get the benefits from AI, without suffering from increased risks and bureaucracy.
It is throwing up countless new questions in the industry, and it is clear there are few easy answers. We look forward to Sheldon’s conclusions.
Written by Jonathan Davidson, Founding Partner, Prysm Global
