This Week in Brief

The story has shifted from what advisers are doing with AI to what their clients are doing with it. New research shows AI has quietly become a normal part of the client journey — and that many clients don’t understand what protection they lose by using it. Alongside that, three separate voices set out practical rules for keeping a firm’s own AI use inside the regulatory perimeter and away from client confidentiality risk.

Key Developments

Clients Are Turning to AI — and Don’t Know What Protects Them

Two pieces of research landed together this week and are best read side by side. Wealthtime, via the lang cat, found 63% of advisers report clients independently using AI in relation to their financial advice, with a further 31% suspecting some clients are doing so — on average, advisers estimate 28% of their client base uses AI to some extent, mostly to prepare questions before meetings (59%) or check advice already given (50%). Separately, the FCA found 80% of 18-40 year-old investors have used AI for investment-related support, and worryingly, 44% wrongly believe AI-generated financial information is regulated, while 38% think it’s fine to base an investment decision on AI output alone. For advisers, the takeaway isn’t to resist this — it’s that client conversations may need to start addressing AI use directly, including what protection genuinely does and doesn’t apply. (FT Adviser: clients using AI, 28 August 2026; FT Adviser: FCA research, 27 August 2026)

Shadow AI: The Confidentiality Risk Inside Your Own Practice

Gareth Fatchett, partner at FS Legal, flagged a gap that’s easy to miss: most firms now have an AI policy, but far fewer have thought through what happens when an adviser pastes confidential client details into a public AI tool to speed up a draft. His point is that this isn’t advisers trying to cut corners — it’s well-intentioned “shadow AI” use that firms often have no visibility of. His practical fix: be explicit about which tools are approved, what information can never be entered into them, and treat any public AI platform as unsuitable for client-identifiable data unless it comes with proper enterprise-grade contractual protections. (Professional Adviser, 25 August 2026)

Where Does Agentic AI Cross Into Regulated Advice?

Chris Davies of Model Office set out where the regulatory line actually sits: the FCA regulates activities, not technologies, so an AI explaining an ISA is fine, but one that analyses a client’s finances and recommends a specific investment is edging into the regulated perimeter — and an agent that can also execute that recommendation is further in still. He ties this to the Advice Guidance Boundary Review’s “targeted support” regime (live since April) and the Mills Review’s new concept of “advice-like support” — AI help that looks like regulated advice to a client without technically sitting inside the perimeter. His argument: RegTech needs to move from sitting around the AI to monitoring inside it, so firms can evidence how an AI-influenced outcome was reached. (Professional Adviser, 26 August 2026)

Update: The “AI for Admin Only” Comfort Line Has a Blind Spot

Last week we covered new GBST/lang cat research showing advisers are comfortable with AI on admin but far more cautious near client money. A follow-up analysis from Resultsense argues that comfort line measures the wrong thing: it tracks how much authority advisers hand to AI, not what happens to an adviser’s own judgement after two years of approving decisions a reliable tool keeps getting right. The argument, drawing on comments from The Oasis Group’s John O’Connell, is that firms should test whether reviewers could still reach an independent view without the AI output in front of them — not just whether they signed it off. Worth a look for any firm treating a low AI-override rate as pure reassurance rather than a risk signal in its own right. (Resultsense, 25 August 2026)

From the Trade Press

  • Perplexity and Flanks partner to bring wealth data into AI workflows — Data infrastructure firm Flanks has partnered with AI search engine Perplexity so advisers can query portfolio holdings, positions and transaction histories directly, feeding into meeting prep, monitoring and reporting. A sign that the “AI agent” pitch to advisers increasingly depends on who controls the underlying data pipes, not just the model. FT Adviser, 26 August 2026
  • A simple ladder for judging any AI output before it reaches a client — Phillip Wickenden proposes thinking about AI use in three tiers: it assists (rewrites material you already own), it informs (supplies a fact or rule you didn’t already know), or it influences (shapes the actual recommendation). The further up the ladder, the more evidence and challenge the output needs before it’s relied on — a useful quick test for any adviser unsure whether a given AI use needs more scrutiny. FT Adviser, 27 August 2026

What to Watch

Keep an eye on how firms respond to the shadow AI issue — expect more practices to publish explicit rules this autumn on which AI tools are approved for client work and which are banned outright. Separately, watch the Advice Guidance Boundary Review and Mills Review’s “advice-like support” concept develop, as it’s likely to shape how far AI tools can go before they need FCA authorisation.

Sources: FT Adviser, Professional Adviser, Resultsense, web search. Compiled 28 August 2026.