AI Adoption Plan: Who’s accountable for agentic commerce?

HM Treasury's Financial Services AI Adoption Plan, published 14 July as part of the Mansion House package, puts the accountability question squarely on the regulator's desk: one of its 10 recommendations calls on the FCA to review how general-purpose AI tools are already delivering financial guidance outside the regulated perimeter. That recommendation draws directly on the Mills Review, an independent review commissioned by the FCA's board earlier this month. Both raise the same pointed question: who actually sits behind the AI making those decisions, and what happens to accountability once an AI interface stands between consumers and their money?

Who sits behind the AI matters

The Mills Review raises a genuine concern: as consumers rely on AI agents to search, compare and transact, whoever controls that AI-mediated interface could capture value and influence choice. That's not really about protecting financial providers' revenue; it's about where accountability sits. Banks and payment providers are regulated: they have obligations to protect customer data, treat customers fairly, and give people a route to complain if something goes wrong. 

General-purpose AI tools like ChatGPT, Claude and Gemini don't sit inside that same regulatory perimeter, even when people are relying on them for financial decisions. The Review found that around a quarter of UK consumers already trust general-purpose AI tools for financial advice, often without realising that the usual routes to complaint and redress won't apply there.

That recommendation goes further than a review: it also calls on industry to agree on a consistent way of disclosing when a service sits inside the regulated perimeter and when it doesn't. This accountability gap is now a government-recommended action, with the FCA and industry both handed something concrete to do about it.

That's precisely why the rails underneath agentic payments matter as much as the interface on top. Regulated banks and payment providers are the trust anchor consumers already have, not a new, unaccountable layer sitting between them and their money. 

This is the same argument that's been building around payment sovereignty more broadly: cards already route the vast majority of UK retail spending (84% by turnover) through a non-domestic duopoly, costing UK businesses roughly £1.5bn a year in fees. Handing the agentic era to the same rails, or to an unaccountable interface layer on top of them, would repeat the mistake rather than fix it. Open banking, built and supervised domestically, is the alternative that keeps trust, accountability and value where consumers can actually see them.

Three things the industry needs to get right

The Mills Review doesn't call for a new regulatory regime. Its preferred path builds trusted agent standards through the existing Open Finance framework, so agents can operate across systems without starting from scratch. We think that's the right call, and it points to three things the industry needs to get moving on:

• Partner with the right regulators early: Standards built with input from the FCA and industry will hold up better than standards built in isolation and presented afterwards.

• Build with guardrails first: Identity, mandate, consent and revocability are the differences between an agent acting on your behalf and one acting instead of you, not features to bolt on later.

• Educate payers and merchants before pushing adoption: Trust in agentic payments will be won or lost on whether people understand what they've authorised and are confident in their decisions, and not on how quickly the technology arrives.

The Mills Review has just given the industry a clear, sensible path to bring agentic payments into that same trusted framework as other payment methods, and we're excited to be a part of the FCA's cohort for AI Live testing to help make agentic commerce a reality.