Building the UK's agentic financial services infrastructure: creating the future, building on the past? Part two

'Originally written for and published by UK Finance'

Existing policy protocols for agentic buildout v. new requirements

The Mills Review and the two HM Treasury publications (AI Adoption Plan and Modernising Payment Services consultation) together reveal a productive tension between existing policy and the need for new concepts and legal boundaries. If the UK is to move beyond incremental common law development to a coherent infrastructure for agentic payments, new technical and policy standards must be developed in tandem with existing liability norms — each informing the other.

The UKJT's Legal Statement underscores why this tension between continuity and novelty matters in practice. Where an AI agent operates as a tool under direct human supervision — analogous to a payment service provider using AI for fraud screening — existing negligence and professional liability principles apply comfortably (i.e. there's a human to take accountability). But where an agent autonomously initiates transactions across multiple counterparties, uncertainty arises. The UKJT lists rapid system operation, speed of functioning and interfaces between human and AI decision-making as likely problem areas, particularly given the opacity of complex models.

On the side of continuity, Mills affirms the robustness of the Senior Managers Regime for AI operating at lower levels of autonomy and finds Consumer Duty capable of governing AI-mediated consumer outcomes.

Mills, though, recommends frameworks for data, identity, authority to act, execution and liability need to operate together as building blocks for agentic finance, noting that "existing payment frameworks assume human approval and are not aligned to delegated or autonomous agent execution".

The HMT Modernising Payments consultation frames this as an invitation by asking industry how existing rules on matters such as authentication, consent and liability will need to adapt. At the same time, the Government explicitly frames the question as a technical "refresh" rather than holistic reform, building on existing and proposed tokenisation, Open Banking, smart data, AI and payments regulation.

Similar to Mills, the AI Adoption Plan calls for entirely new frameworks, this time relating specifically to Know Your Agent protocols and interoperable machine-to-machine authentication standards — neither of which has any precedent in UK financial regulation. 

The message is clear: where existing protocols can be extended, they should be. Where they cannot, industry and regulators must build new landscapes together, as they have done on many previous occasions. The commercial model imperative though may well be easier to fulfil this time round.

Next steps for the UK's financial sector

Stakeholders have a narrow window to shape this pivotal moment for true agentic commerce at scale. Priority actions include:

• Consider the questions in the HMT consultation. Questions 15 and 16 ask directly how payment services regulation should adapt for agentic payments and what other innovations the Government should account for. This is a once-in-a-generation opportunity to shape the legislative framework — and the consultation period closes on 6 October 2026.
• Prepare evidence for the FCA's perimeter review. The AI Adoption Plan recommends a comprehensive FCA assessment of LLM-generated financial guidance and its consumer and competitive impacts. Stakeholders should begin gathering data now on how unregulated AI outputs are affecting behaviour and market dynamics.
• Join the trust framework coalition. The AI Adoption Plan envisages industry-led development of agentic payments standards — covering liability, Know Your Agent protocols, and machine-to-machine authentication — potentially convened by CFIT through its coalition model. Interested players should step forward to join the conversation and ensure that their technical and commercial leads are playing an active part.
• Test agentic capabilities through the FCA AI Lab. Developers of agentic products should use these facilities to accelerate safe deployment and feed real-world insights back into the standard-setting process.

The alignment between these three reports is not accidental. It reflects a maturing UK policy consensus that agentic finance is both inevitable and desirable — but only if the infrastructure is built deliberately, collaboratively, and with consumers at its centre. The window to shape that infrastructure is now fully open.

Access part one here.

Authors: Fiona Ghosh, Partner and Patricia Wade, Expertise Counsel.