The FCA's call for expanded supervisory powers over autonomous AI models reflects a real anxiety: 40% of Britons now use chatbots for financial guidance, many of them unvetted by any regulated firm. The regulator wants to extend its perimeter to capture AI-powered advice tools and build a public-interest financial guidance service. For mid-market law firms, insurers, accountancies and financial services providers, this means one thing clearly: if you deploy AI to give advice or make decisions that touch regulated activity, the FCA—and the SRA, PRA, ICO and FRC depending on your licence—now sees you as accountable. This is not a future risk. This is current law under the Consumer Duty (PS22/9) and soon to be reinforced by the UK AI Act's own mandatory controls for high-risk systems.

What the FCA's statement really reveals is that the industry has a governance crisis, not a technology crisis. Firms are racing to adopt AI—often using off-the-shelf large language models like ChatGPT, Claude or Microsoft Copilot without any firm-specific safeguards, validation protocols or audit capabilities. They are treating AI as a productivity tool, not as a regulated activity. Meanwhile, unregulated third parties are openly offering AI-powered financial advice to consumers with zero compliance infrastructure. The FCA's worry is understandable. But the regulator is focusing on the visible symptom (consumer-facing rogue bots) when the real disease is inside your own four walls: staff using untrained, unmonitored AI to draft advice, make underwriting decisions or process claims.

Here is the hard truth: generic large language models cannot be safely deployed in regulated advisory or decision-making contexts without a robust, audit-trail-enabled wrapper that includes human review, explainability, bias detection and documented sign-off. Trovix's approach—built from the ground up for regulated financial services and legal practice—uses retrieval-augmented generation (RAG) architecture combined with document-intelligence and intake automation to ensure that every AI output is traceable, contestable and tied to firm-specific knowledge and governance rules. This is why Trovix Aria is not a chatbot: it's a fee-earner assistant with built-in compliance oversight. Tools like Harvey or Legora take similar discipline to legal work. But many firms are still downloading free versions of ChatGPT and hoping for the best. That approach will not survive the FCA's next enforcement round.

What you should do now: First, audit every instance of AI use in your firm—be honest about what is happening in spreadsheets, emails and unlogged prompts. Second, map which AI activities touch regulated advisory, underwriting, claims assessment or investment decisions. Third, if they do, you need documented governance: a data lineage, a human review stage, an audit trail and a clear rationale for why this AI use complies with Consumer Duty (PS22/9), the UK AI Act's proportionate-risk framework, and your sectoral regulator's expectations (PRA SS1/23, FRC ISA UK 260, SRA Code). Fourth, deploy AI platforms designed for regulated practice—not consumer tools retrofitted with disclaimers. The cost of doing this properly is far lower than the cost of an FCA investigation or a Consumer Duty breach fine.

Source: City AM