The FCA has not written rules specifically for AI agents that make payments, and as of November 2026 it does not plan to. Instead, it is working with HM Treasury on how existing payment rules should adapt, while its Mills Review chose to rely on the Consumer Duty rather than new AI rules. For consumers, the key point is a gap: the FCA's protections reach your bank and payment provider, but not a general-purpose AI agent that is not an authorised firm.

This explainer summarises regulator publications and legal commentary. It is not financial or legal advice.

What has the FCA actually said?

  1. FCA 2026 payments priorities commit to work with HM Treasury on agentic payments
  2. FCA Mills Review declines to recommend AI-specific rules
  3. HM Treasury AI Adoption Plan makes agentic payments a priority
  4. CMA, FCA, ICO and Ofcom joint statement on personal AI agents
  5. Treasury payments consultation on a trust framework still awaited

2026 payments priorities (March 2026). The FCA's annual payments priorities included agentic AI for the first time, committing it to work with HM Treasury on how agent-initiated payments fit the rules (Linklaters, Payment Expert).

The Mills Review (6 July 2026). The FCA's review of AI in retail financial services declined to recommend AI-specific rules, concluding that existing regimes, led by the Consumer Duty, were the right starting point (Freeths, Deloitte). It also observed that the payments rulebook "assumes a human approves each transaction" (GoCardless). The FCA published it on 6 July 2026.

The joint regulator statement (30 October 2026). Through the DRCF, the FCA joined the CMA, ICO and Ofcom in saying "existing legal and regulatory frameworks apply to personal AI agents", and that they "will not hesitate to use our powers" (Resultsense). See our DRCF statement explainer.

What is HM Treasury planning?

HM Treasury's Financial Services AI Adoption Plan, published on 14 July 2026, lists agentic payments as a priority. According to GoCardless, a forthcoming payments consultation will develop a "trust framework" covering:

  • Liability: who pays when an agent-initiated payment goes wrong.
  • Know Your Agent: verifying that an agent is genuine and authorised by the customer, in the same way firms verify customers today.
  • Machine-to-machine authentication: how an agent proves its authority to a payment system without a human tapping "approve" each time.

We found no published date for that consultation as of November 2026.

Where the Consumer Duty applies, and where it does not

Situation (as of November 2026) Who is responsible? Consumer Duty applies?
Your bank's own AI assistant (e.g. Starling, Revolut) The bank, an authorised firm Yes
A payment through Visa or Mastercard agent credentials Your card issuer for the card side Yes, for the issuer's conduct
A general personal agent (dots, Muse, Duet) buying for you The AI vendor and the retailer Not directly; vendor usually not FCA-authorised
A self-hosted agent (OpenClaw) using your card Largely you No

Sources: Freeths, Deloitte.

The Consumer Duty requires authorised firms to deliver good outcomes for retail customers. If your bank deploys an AI assistant, that assistant's conduct is the bank's responsibility under the Duty. Freeths and Deloitte both note that the Duty does not reach general-purpose AI agents run by firms that are not authorised, which is how most personal agents are offered.

Who is liable when an agent pays wrongly?

The Payments Association describes liability for agentic commerce in UK retail as an "unresolved" question. Several layers interact:

  • The retailer is still bound by the Consumer Rights Act 2015 and, for online purchases, the 14-day cancellation right.
  • The CMA says businesses answer for their AI agents "as they do for their employees". That applies to retailers and agent providers dealing with consumers (CMA guidance).
  • Your card provider may offer chargeback or, for credit cards, Section 75 protection, though how these apply to agent purchases is untested. See Section 75 and AI agent purchases.
  • Payment services rules on unauthorised transactions were written for humans approving payments, which is why the Treasury wants a trust framework.

How the industry is filling the gap

While policy catches up, card networks and agent vendors have built their own safeguards:

Pros

  • Visa and Mastercard issue agent-specific tokens limited by merchant, amount and time.
  • Claude Duet and Meta Muse use single-use cards limited to one purchase.
  • OpenAI dots always returns money transfers to the user.

Cons

  • These are commercial safeguards, not statutory rights.
  • Protections differ between vendors and can change without consultation.
  • None has been tested by a UK court or the Financial Ombudsman as far as we found.

Our guide to Visa and Mastercard agent payments in the UK explains the card-network side, and connecting AI agents to UK banks covers Open Banking.

What should UK consumers do now?

  1. Keep approvals onRequire your agent to ask before every payment.
  2. Pay by credit card for larger purchasesSection 75 may help on purchases over £100 and up to £30,000, and chargeback on smaller amounts.
  3. Prefer authorised firmsA bank's own assistant brings the Consumer Duty with it.
  4. Keep recordsScreenshot the agent's log, the approval and the order confirmation.
  5. Report problems fastTell your card issuer immediately if a payment looks wrong.

Bottom line

The FCA's position is that existing rules, led by the Consumer Duty, are enough for now, while HM Treasury works on a trust framework for agent payments. That leaves a real gap for general-purpose personal agents run by firms that are not authorised. Until the consultation lands, your best protection is to keep a human approval on every payment and pay in a way that preserves your card rights. The US hub explains how AI agent payments work technically.