Agentic AI and the trust gap regulators haven’t caught up to yet

On 14 July 2026, HM Treasury published its Financial Services AI Adoption Plan, prepared by a group of independent ‘AI Champions’ brought in to work out what is actually stopping UK financial services from scaling AI safely. Most of the plan reads as you would expect: regulatory clarity, skills gaps, cloud dependency. One section reads differently, because it comes down to the fact that nobody has yet worked out who is responsible when an AI agent spends your money on your behalf.

The plan calls this ‘agentic payments readiness’, and its own language is unusually candid. Firms have highlighted significant uncertainty around legal and regulatory accountability, including liability and consent, alongside heightened concerns about the potential for fraud within automated payment flows’.

A trust framework built on three pillars

The Treasury’s proposed solution is to build a ‘trust framework’ for agentic payments around three things. Legal and liability frameworks, so it is clear who is accountable when an autonomous agent transacts. Authentication and governance, interoperable technical standards for safe machine-to-machine transactions. And Know Your Agent: a direct echo of Know Your Customer (KYC), a standardised way of identifying and verifying the AI agent doing the transacting, not just the person who switched it on.

KYC exists because a bank needs to know who it is actually dealing with before it lets money move. KYA is the same instinct applied to a world where the thing initiating the payment might not be a person at all, and the Treasury is openly saying this framework does not exist yet.

People are already trusting AI faster than the rules can catch up

The plan cites the FCA’s own Mills Review on AI and retail financial services, and the numbers are striking. One in five UK adults are already open to AI making financial decisions for them, and the appetite is strongest exactly where the stakes are highest: debt advice, pensions, investments. Around 26% already trust general-purpose tools like ChatGPT or Gemini for financial advice, and the review notes most of them have limited awareness that formal routes to recourse simply will not apply if that advice turns out to be wrong.

This is a trust gap. Adoption is running ahead of protection, and the products people are trusting were never built – or regulated – for the job they are quietly being asked to do.

Why this is a design problem, not just a policy one

This is the same underlying question we wrote about a few weeks ago, in relation to the FDA’s discussion paper on generative AI medical devices: how directive is this output, and does the product make that level of directiveness honest to the person relying on it? Two regulators, two industries, both openly working out how to hold an autonomous system accountable within weeks of each other.

Know Your Agent is a genuinely interesting design brief. What does it actually look like, inside a product, to show a customer which of their agents is transacting, under what authority, within what limits in a way they can verify at a glance rather than simply trust? We have already had to answer a version of this question. For FDB’s OptimiseRx, the task was making sure a clinician could tell, instantly, how much weight to put behind a piece of guidance before acting on it. Agentic payments need the equivalent for money: a customer who can see, at a glance, exactly what an agent has been authorised to do on their behalf, and trust that boundary is real.

The Treasury’s consultation on modernising payment services regulation is now open, which means the rules for this are still being written – and the products that will have to live inside them are being designed right now, ahead of the framework, the way they always are. ‘Know Your Agent’ will become a real screen, a real interaction, a real moment where a person decides whether to trust what they are seeing. Someone has to design that moment well.

That is the part we find genuinely interesting, and the part we stake our business on. If you are building in this direction and want to think through what verifiable trust actually looks like inside your product, we would like to talk.