The FCA's July call for expanded regulatory powers over AI-powered financial advice responds to a genuine shift: 11 million UK adults are already willing to let algorithms make money decisions, and nearly a third would trust chatbots with pensions. This is not hypothetical risk. It is live consumer behaviour outpacing the rulebook. Under FCA Consumer Duty PS22/9, firms remain liable for the advice their AI systems give, even when those systems operate within 'agreed parameters'. The problem is not whether the FCA will regulate AI — it will. The problem is whether mid-market advisers, wealth managers and platforms have the technical infrastructure to prove they comply when they are audited.

This story sits within a larger pattern. Regulators globally — the EU AI Act's mandatory conformity assessments, the Lloyd's Blueprint Two's ESG-plus governance requirements, ISO 42001 certification creeping into insurance procurement — are all moving toward the same assumption: AI systems must be inspectable and explainable at the point of decision. Yet most financial services firms deploying chatbots today use off-the-shelf large language models from Microsoft Copilot or Claude, fine-tuned with proprietary data but still fundamentally opaque. When an FCA examiner asks 'why did your AI recommend this pension to this customer?', the honest answer is often 'we don't know'. That is not a regulatory problem yet. It will be.

Trovix's view is straightforward: compliance begins with auditability, not permission. The FCA will almost certainly introduce a regulatory perimeter expansion by 2027 that requires firms to demonstrate human oversight, algorithmic bias testing, and decision traceability for any AI system that advises on regulated products. Generic chatbots do not meet that bar. They are good at sounding confident. They are poor at proving they are right. This is why systems built on retrieval-augmented generation (RAG) architecture — where the AI cites specific policy documents, precedents or compliance rules before making a recommendation — outperform black-box LLMs in regulated environments. Trovix Reach embeds regulatory guardrails into client-facing AI precisely because it knows which rules matter and can show its working. Systems like Harvey or Legora do similar work in legal and insurance claims, but financial advice is harder: the liability tail is longer and the regulatory perimeter is wider.

What should a mid-market wealth manager, insurance intermediary or fintech do right now? First, audit your current AI chatbots. If they cannot explain their advice by reference to documented policy, guidance or precedent, they are regulatory liabilities, not assets. Second, begin building an AI governance dashboard — Trovix Audit does this — that logs every material decision your systems make and flags ones that breach compliance rules before they reach clients. Third, plan for the FCA's new rules as if they are already law. The regulator does not usually issue guidance and then fail to enforce it. Firms that wait for the formal rules will find themselves mid-compliance rebuild when examination letters arrive. Firms that build auditability into their AI now will find the new rules merely formalize what they already do.

Source: City AM