NatWest Retail customers say AI oversight is important

NatWest has published research finding that three in four of its Retail customers — more than 2,400 people surveyed, not a poll of the UK public — said strong, independent regulatory oversight of artificial intelligence in banking was important. NatWest separately benchmarked its findings against a nationally representative sample of 1,800 UK consumers, but the material it has published does not show that the 75% figure, or the other headline percentages, came from that wider sample rather than its own customer base. The finding is solid evidence about what NatWest's customers say; it should not be read as evidence about UK consumers as a whole.

The same customer sample found that 81% identified the ability to reach a real person when needed as the single most important factor in trusting AI in financial services.

Set against this is the Financial Conduct Authority's Mills Review, published on 6 July 2026, which examined how AI could reshape UK retail financial services and concluded that adoption will hinge on trust, control and access. The review recommended firmer system-wide coordination and oversight of AI, but the FCA has not introduced a separate AI rulebook for regulated firms. For UK banks rolling out AI-assisted customer service, the question is not whether customers like AI — many already use it — but whether banks can show the safeguards, disclosure and human escalation that both their customers and the regulator's research say matter.

What NatWest actually found

NatWest's AI Adoption Report draws on feedback from more than 2,400 NatWest Retail customers, benchmarked against the separate 1,800-person nationally representative sample. The customer-level findings, as published, are:

NatWest has not published the survey's fieldwork dates, research agency, sampling method, weighting or exact unrounded base, and it has not published a results table separating the customer sample from the national benchmark. The figures above describe NatWest's own customers, not a national statistic.

Trust does not mean rejection of AI

The same customers report substantial existing AI use: 71% said they used AI tools at least occasionally, and 40% said they used them most days or every day. Seventy-four per cent said AI simplifies choices, and 22% already use AI platforms for financial information and advice. Sixty-four per cent said they would be interested in AI guiding them through a mortgage application and identifying suitable options — a product area where a wrong recommendation matters, since a mortgage is a large, long-term secured debt.

NatWest points to two live examples: an audio-visual Spending Insights tool trialled through Royal Bank of Scotland, and a Fraud Triage Agent built into its Cora digital assistant. Together, the figures describe conditional acceptance rather than resistance: customers want AI's convenience alongside disclosure, reliability, data clarity, an accountable oversight body, and a way to reach a person.

What "human support" means in practice

Wanting access to a person is not the same as wanting a person to handle every interaction. NatWest's published finding is limited to the fact that 81% identified access to a person when needed as their most important trust factor; the underlying questionnaire wording has not been published, so no further conclusion should be drawn about what respondents were choosing between.

UK regulation already engages with this question. The FCA's Consumer Duty, which applies to open retail products and services from 31 July 2023 and to closed products and services from 31 July 2024, requires firms to provide support that meets retail customers' needs, including the needs of customers with characteristics of vulnerability, and to avoid unreasonable barriers across the product lifecycle. The Duty is channel-neutral: it does not require every interaction to include a human being.

FCA guidance to retail banks and building societies is more specific. The regulator does not prescribe that every support channel must include a person, but it has said a real-time human interface is likely to be needed for some non-standard, complex or sensitive issues, including fraud, security problems, account-access failures and bereavement-related journeys. Separate FCA good-practice material gives an example of good practice in which a sensitive bereavement-related chatbot query is routed to a human customer-support representative.

The Information Commissioner's Office adds a data-protection layer: organisations using AI-assisted decisions involving personal data should be transparent and accountable, explain decisions meaningfully, and provide a capable human contact through whom an affected person can query or contest a decision. That guidance is under review following the Data (Use and Access) Act 2025, and the precise legal position depends on whether a decision is solely automated and produces a legal or similarly significant effect.

The UK regulatory position

The FCA's stated approach, published on 8 September 2025 and last updated on 13 February 2026, is principles-based and outcomes-focused rather than built around AI-specific rules. It says it does not currently plan additional AI-specific regulations for regulated firms; instead it relies on existing frameworks, including the Consumer Duty and senior-management accountability, to hold firms to account for the outcomes their AI systems produce. For firms it regulates, that reliance on existing frameworks means AI-assisted banking services remain subject to conduct, governance and accountability requirements even without a dedicated AI rulebook.

Separate research commissioned by the Bank of England and the FCA in 2024 found that 75% of responding UK financial-services firms already used AI, with a further 10% planning to within three years. Eighty-four per cent had assigned a person accountable for their AI use, but only 34% reported a complete understanding of the AI systems they used, with 46% reporting only partial understanding. That gap between assigned accountability and demonstrated understanding is a governance challenge in its own right; the survey material does not show whether it is what NatWest's customers meant by independent oversight.

The Mills Review: what changes, and what does not

The FCA published the Mills Review on 6 July 2026, drawing on a survey of more than 5,000 UK financial-services consumers, FCA-led focus groups, stakeholder submissions and analysis of international approaches. It concluded that retail finance is moving towards AI-enabled, continuous and delegated services, that adoption will depend on trust, control and access, and that one in five UK adults is already open to AI making financial decisions for them.

The review set out seven recommendations, including securing and adapting the FCA's regulatory perimeter, strengthening system-wide coordination and oversight, and building an AI-enabled supervisory model. These are recommendations to the FCA Board, not new rules, and the material reviewed does not establish which had been formally accepted or implemented by 26 September 2026. Current rules remain the operative basis for supervision now; the recommendations describe work the FCA may take forward, including further scrutiny of AI activity that sits outside its regulatory perimeter.

The perimeter problem

That perimeter question affects what protection a customer actually has. A general-purpose AI chatbot is not regulated by the FCA merely because it produces financial information. An AI tool specifically configured to provide regulated financial advice would be likely to fall within the FCA's remit, but whether protection applies, and to what extent, depends on the firm, the activity performed and the circumstances of the case.

A separate FCA survey of 666 UK respondents aged 18–40 who owned or were considering investments, conducted on 24 July 2026 and published on 27 August 2026, found that 56% trusted AI tools for investment purposes, but 44% mistakenly believed AI-generated financial information was regulated, and 32% wrongly expected Financial Services Compensation Scheme or Financial Ombudsman Service protection if AI advice went wrong. This is a narrow younger-investor sample, not representative of all UK adults, but it points to a real risk: assuming FOS or FSCS protection applies where it does not.

What to watch next

The FCA has not published a timetable for which Mills Review recommendations it will take forward, or when. Readers who want the underlying detail should go to the FCA's published Mills Review and its "AI and the FCA: our approach" page directly, and to NatWest's own AI Adoption Report for the customer research, rather than to a commercial intermediary. Worth watching: any FCA review of general-purpose AI tools outside its perimeter, further FCA good-and-poor-practice publications on AI and consumer support, and any update to the ICO's automated-decision guidance following the Data (Use and Access) Act 2025.