The FCA has stopped watching from the sidelines. On 6 July, the watchdog signalled that it will no longer tolerate unregulated AI systems operating in the retail financial advice space—and it wants new powers to enforce this. The headline is clear: as Brits increasingly ask ChatGPT, Claude and home-grown chatbots for money guidance, the FCA sees a gap in its rulebook. What matters for your firm is concrete: if you deploy AI in any customer-facing financial context, you now face a regulatory environment that is shifting from permissive to prescriptive. The days of treating AI as a test-and-learn playground are ending.

This is part of a wider pattern. The FCA's position in 2023—which amounted to 'let's see what happens'—has given way to the same institutional clarity we saw from the SRA on legal AI (SRA Code Practice Note guidance, 2024), the ICO on GenAI and data governance, and the EU's approach under the AI Act. Regulators globally have watched AI vendors ship products faster than governance structures could be built, then scrambled to contain the damage. The Financial Conduct Authority Consumer Duty (PS22/9) already mandates that you act in the customer's best interests; AI agents that give bad advice or operate in shadow regulatory space violate that duty. The FCA is now saying: we will define what proper AI governance looks like, and it will not be optional.

Here is Trovix's view, plainly stated: most off-the-shelf large language models—including the latest versions of ChatGPT, Claude and Microsoft Copilot—are not fit for regulated financial advice without significant wrapping, guardrails and governance. They hallucinate, they go out of date, and they lack the audit trail that FCA Consumer Duty and PRA SS1/23 demand. We see firms deploying Harvey or Legora for legal work because those tools were built inside regulatory constraints from day one. In financial services, the equivalent rigour barely exists in the market yet. This is why firms should not bolt a chatbot onto their website and call it AI governance. You need three things: (1) a model trained on or constrained to your licensed advice framework; (2) a full audit and decision trail that a regulator can inspect; (3) clear human oversight at the point where advice becomes binding. Most existing AI products skip one or more of these.

What should a mid-market financial services firm, accountancy practice or legal firm do today? First: audit your current AI use. If you have a generalist chatbot advising customers, stop it or rebuild it. Second: expect the FCA's formal guidance (now in consultation, likely to be final by Q1 2027) to require you to document your AI governance—similar to how ISA UK (FRC) frameworks demand you record audit decision-making. You will need Trovix Audit or an equivalent to show your work. Third: use Trovix Watch to track the FCA's own AI rulebook as it emerges; this will evolve rapidly. Finally, if you are planning to deploy a client-facing AI assistant, insist that your vendor can prove compliance with the emerging standard before you go live. The cost of a compliant build now is far less than the cost of a regulatory enforcement action later.

Source: City AM