AI Regulation Tracker / United Kingdom
UK Treasury Backs Industry AI Adoption Plan for Financial Services
On July 14, 2026, HM Treasury published the industry-led "AI Adoption Plan: Financial Services" and the government welcomed the plan and accepted its recommendations. This is a non-binding plan written by two independent AI champions, not a rule. It changes no legal duty, but it points to where the FCA and PRA are likely to work next.
On July 14, 2026, HM Treasury published the "AI Adoption Plan: Financial Services." It landed alongside the Chancellor's Mansion House speech, delivered the same evening at the government's AI Adoption Summit for the sector. The publication page states it plainly: the FS AI Champions, Harriet Rees and Dr Rohit Dhawan, have developed an AI adoption plan for the UK's financial services sector. The government welcomed that plan and accepted its recommendations.
Two names matter here. Harriet Rees is Group Chief Information Officer at Starling Bank. Dr Rohit Dhawan is Head of AI and Advanced Analytics at Lloyds Banking Group. They were appointed as independent Financial Services AI Champions, and the plan is their work. That authorship is the whole point of the story, so I want to be careful with it before we go any further.
What this actually is
This is an industry-led plan that the government has accepted. It is not legislation. It is not a new rule. It is not an FCA or PRA requirement that has taken effect. Nothing in the legal duties of a UK bank, insurer, asset manager, or payments firm changed on July 14 because this plan was published.
The recommendations belong to the AI Champions, not to the FCA. It would be easy to read a government summit and a Chancellor's speech and conclude that a regulator has just handed down expectations. That is not what happened. Two independent industry figures wrote a plan, and HM Treasury accepted it. The accepting is meaningful, because it commits the government to act on the recommendations and directs future work, but the plan itself is a set of recommendations, not a rulebook.
What the acceptance does is set direction. When the Treasury publishes an adoption plan and says it accepts the recommendations, it is telling the market where government, the FCA, and the PRA intend to focus next. It is a signpost for forthcoming workstreams, not a wind you are legally required to turn into today.
What the plan sets out to do
The plan contains roughly ten recommendations aimed at accelerating safe AI adoption across UK financial services. At a summary level, they span several areas. There is the regulatory framework itself, and how it should treat AI. There is AI-powered financial advice, which sits close to a long-running question about where advice ends and guidance begins. There is the regulatory perimeter, meaning what falls inside and outside supervision as AI reshapes services. There is operational resilience, the discipline of keeping critical services running when the technology underneath them is new. There is skills and talent, addressed through a proposed City skills compact. And there is agentic payments, the emerging world of AI systems that can initiate transactions on a customer's behalf.
Read together, these recommendations signal near-term workstreams for the FCA and the PRA. Expect attention to human-in-the-loop expectations, to audit trails that let a firm show how an AI-assisted decision was reached, to the treatment of agentic finance, and to the advice and guidance boundary. None of that is a live requirement yet. It is a map of where the regulators are being pointed.
Because the full ten-recommendation text is detailed and I am describing it at a summary level, I want to avoid overstating any single item. The safe reading is the shape of the plan, not a line-by-line list of duties, because there are no new duties here to list.
What this is not
I want to be precise, because the framing invites over-reading.
This plan does not impose a disclosure obligation, a control requirement, or a filing duty on anyone. It does not switch on an FCA or PRA rule. It does not set a binding deadline. It does not attribute any expectation to the FCA as a matter of current regulation, because the recommendations are the AI Champions' plan that the Treasury accepted, not FCA policy that has taken effect.
What it does is real all the same. Government acceptance of a plan is how the UK often begins a regulatory workstream. The value in reading it now is anticipation. You get to see the topics your regulators will likely take up, and you get to prepare your own governance before anything hardens into a requirement.
What this means for finance teams and vendors
For UK-authorised firms, the plan is a preview of the agenda. If you are already thinking about human-in-the-loop controls, audit trails for AI-assisted decisions, resilience of AI-dependent services, and the boundary between advice and guidance, you are working on the same questions the plan foregrounds. If you are not, this is a clear, early signal of where scrutiny is heading, especially on agentic payments and AI-powered advice.
For US CPAs and finance leaders whose firms operate in the UK, and for US AI vendors selling into UK financial services, the plan is a non-statutory roadmap. The FCA and PRA are not your regulators for US filings, and this plan carries no authority over anything you do in the United States. But the pattern it describes travels. US parents of UK-authorised banks, insurers, asset managers, and payments and fintech firms now have a clearer view of the expectations their UK operations are likely to face. US vendors get an early read on what UK buyers will start asking for, particularly around agentic payments and the ability to evidence how an AI system reached a decision.
The practical thread is governance. The plan does not tell you to build controls. It tells you which controls the UK is about to care about. Data provenance, human accountability for AI-assisted output, auditability, and resilience are the themes running through it, and those are the same themes that turn into supervisory findings when they are ignored.
What to do now
Read the plan for what it is, an accepted set of industry recommendations that directs future work, not a rule that has taken effect. Map where AI already touches regulated activity in your firm, with particular attention to any payment initiation, any advice or guidance boundary, and any AI-assisted decision that a supervisor might one day want explained. Keep a named person accountable for AI-assisted output and make sure you can show your audit trail. Watch for the FCA and PRA workstreams the plan signals, since that is where actual requirements will eventually come from. And do not change a control or a disclosure on the strength of this plan alone. If you act, base it on your own applicable rules and professional advice, not on a UK adoption plan.
Questions professionals are asking
Did the UK government issue a new AI rule for financial services?
No. On July 14, 2026, HM Treasury published an industry-led "AI Adoption Plan: Financial Services," developed by the two independent FS AI Champions, and the government accepted its recommendations. It is a plan, not a statute, a rule, or an FCA or PRA requirement, and it changes no legal duty on its own.
Who actually wrote the recommendations?
The plan was developed by the two independent Financial Services AI Champions, Harriet Rees, Group CIO of Starling Bank, and Dr Rohit Dhawan, Head of AI and Advanced Analytics at Lloyds Banking Group. The recommendations are their work, accepted by HM Treasury. They are not FCA policy that has taken effect.
What does the plan cover?
At a summary level, roughly ten recommendations spanning the regulatory framework, AI-powered financial advice, the regulatory perimeter, operational resilience, skills and talent through a City skills compact, and agentic payments. It signals near-term work by the FCA and PRA on human-in-the-loop expectations, audit trails, agentic finance, and the advice and guidance boundary.
Does this affect US firms or US vendors?
Not directly. The FCA and PRA do not regulate US filings, and this plan carries no authority in the United States. But US parents of UK-authorised firms and US AI vendors selling into UK financial services get an early, non-binding roadmap of the expectations their UK operations and buyers are likely to face, especially on agentic payments and AI-powered advice.
Should we change our AI controls because of this plan?
Not on the strength of this plan alone. It is a signal of where UK regulation is heading, useful for anticipating FCA and PRA workstreams and getting your governance ready. Any change to your controls or disclosures should rest on your own applicable rules and professional advice, not on an accepted adoption plan.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal, accounting, or audit advice. Confirm how any standard or requirement applies to your situation with qualified professionals in the relevant jurisdiction.