The FCA's call for expanded regulatory powers is not bureaucratic overreach—it is a clear signal that the current rulebook has been outrun. When two in five UK consumers are now turning to chatbots for money advice, and those chatbots operate outside any formal perimeter, you have a massive gap between actual consumer behaviour and actual regulation. The FCA wants to widen its remit to cover autonomous AI models and establish a public-interest AI guidance service. This matters enormously to mid-market law firms, insurers, and accountancy practices. If you are advising financial services clients, or if you yourself operate within the FCA's remit, the regulatory landscape just got tighter. The Consumer Duty (PS22/9) already demands that firms put consumer interests at the centre of governance—AI implementation is now a direct Consumer Duty issue.

This story reveals two things the industry has been slow to admit. First: the distinction between 'AI-assisted advice' and 'AI-generated advice' is collapsing in practice. Consumers do not distinguish between a chatbot trained on financial principles and a qualified adviser; regulators now will. Second: the idea that you can deploy large language models like ChatGPT or Claude into advisory workflows without explicit governance and audit trails is finished. The firms that have treated tools like Microsoft Copilot as productivity gains without rebuilding their control frameworks are now regulatory liabilities. The FCA's push for 'oversight of autonomous AI models' is code for: show us the decision logs, show us the training data, show us the guardrails, show us the human review points.

Here is Trovix's honest take: most financial services firms are implementing AI backwards. They buy a tool—Harvey for legal research, Luminance for due diligence, a generic LLM for client intake—and then layer compliance on top of it afterward. That is backwards. The right approach, which we have embedded into Trovix Aria, starts with the regulatory requirement and the client outcome, then builds the AI around it. With Aria, fee-earners get real-time retrieval of firm knowledge and regulatory standards: that is not a productivity hack, it is a documented audit trail. When you query Aria, the system logs which sources it consulted, what it returned, and what the user did with that output. That is not overhead—that is compliance by design. Firms that have bolted compliance onto generic LLMs after the fact are now scrambling to build those audit trails retroactively. The FCA will not wait.

What should a mid-market financial services firm or an accountancy practice do right now? First: audit which AI tools your teams are actually using. Not the ones you approved—the ones they are using. Second: assess which of those tools touch regulated advice or client guidance. Third: for those tools, build explicit governance: who can use it, for what purpose, with what human review, and what gets logged. Fourth: use regulatory monitoring tools like Trovix Watch to stay ahead of the FCA's detailed guidance as it emerges over the next 12 months. The agency's recommendation for a 'public-interest AI financial guidance service' will come with technical standards and perimeter definitions. You need to know what those are the day they are published, not three months later. Do not wait for enforcement to force this conversation.

Source: City AM