The FCA's July 2026 call for expanded regulatory powers is not surprising; it is overdue. When 11 million UK adults would let AI make autonomous financial decisions, and a third are considering it for pensions and investments, the regulator has a genuine systemic risk problem. But here is what matters for mid-market firms: the FCA's expanded regulatory perimeter will not arrive as a single set of rules. It will come piecemeal through enforcement action, consultation papers, and updated supervisory expectations. By the time formal guidance lands — expect 2027 or 2028 — the firms that have waited for clarity will be scrambling to audit and remediate AI systems already in live use. The Consumer Duty (PS22/9) already requires you to act in customers' best interests; the FCA will simply argue that deploying untested or un-audited AI violates that duty.
This story is part of a broader pattern: regulators everywhere are discovering that AI regulation cannot be written as prescriptive rules first. It must follow from use cases. The EU AI Act splits AI into risk tiers; the ICO's UK GDPR guidance on automated decision-making is now two years behind the actual deployment rate. Insurance firms using AI for underwriting, law firms using AI for contract review, accountancy practices using AI for tax research — none of these sectors had clear regulatory guidance when they started. Each has discovered, in live traffic, that 'AI' is not one thing. A retrieval-augmented generation (RAG) assistant like Trovix Aria that feeds client data into a foundation model behaves differently, and poses different risks, than a fully autonomous agent that makes decisions without human review. The FCA's coming rules will reflect that distinction. Many firms will not.
Trovix's view is blunt: firms that treat AI as a cost-cutting tool rather than a control-point will pay compliance costs later. The industry is littered with examples of well-intentioned AI deployments that failed on governance, not capability. ChatGPT wrapped in a simple chatbot interface looks like advice; it is pattern completion. Harvey and Luminance built for professional services because they understood that lawyers and accountants need explainability and audit trails, not speed at the expense of liability. Microsoft Copilot in Microsoft 365 is different — it is integration within a known ecosystem. The gap between these approaches matters. When the FCA investigates why a firm gave bad AI-assisted advice, regulators will not care that the AI was 'accurate' by some benchmark. They will care whether the firm knew what the AI was doing, whether humans stayed in the loop, and whether decision-making was auditable. Trovix Watch can help you track incoming FCA expectations in real time; Trovix Aria is built on that principle — it supplements human judgment, not replaces it.
What should a mid-market practice do now? First, audit what AI is already in use. If you have a chatbot answering client questions about pensions, investments, underwriting, or tax strategy, you need to know its training data, its failure modes, and what percentage of its outputs you are actually checking before they reach clients. Second, map the ownership: who is accountable if that AI gives wrong advice? Third, start planning for the regulatory perimeter change as though it were already law. The FCA has signalled that autonomous AI models and public-interest guidance services will fall within its perimeter. If your firm is using either, assume you will need to demonstrate human review, audit trails, and testing before the regulator asks. Use Trovix Watch to flag incoming FCA consultations and precedent-setting enforcement actions in real time. The firms that survive the coming tightening will not be those that move slowest — they will be those that understood the risk early and built governance before regulation forced it.
Source: City AM