Since 7 October 2024, banks, building societies and payment firms have reimbursed victims of authorised push payment scams sent over Faster Payments, with the cost split between the sending and the receiving institution. The rule was the first of its kind anywhere. It was also, usefully, a natural experiment, one dated intervention applied to a single payment rail, leaving a clean before and after for anyone willing to measure it.
Frontier Economics measured it. The consultancy’s evaluation, commissioned by the Payment Systems Regulator and published on 1 July 2026, is the first independent assessment of what happens when the obligation to absorb a fraud loss sits with the firms best placed to stop it. The answer turns out to be favourable, and its usefulness reaches well past payments into the far larger question of who answers for money moved by software.
Proof over prediction
Fraud losses sent over Faster Payments fell by around 21% after the requirement took effect, worth roughly £73m a year, with nearly 35,000 fewer individual scams. The figures describe money that stayed in consumer accounts and never reached criminal hands at all. Reimbursement returns funds after the event; prevention keeps them out of circulation entirely, which is the harder outcome and by some distance the more valuable one.
Frontier identified no evidence of firms leaving the market, and none of consumers behaving carelessly on the assumption that someone else would carry the loss. The largest improvements came from the firms with the highest fraud levels beforehand, which is the clearest available sign that a well-set incentive achieves more than enforcement volume ever could.
A policy that pays
Reimbursement rates across all reported claims rose from 54% to 65%. For claims falling inside the policy’s scope, firms now reimburse in 97% of cases. The distance between the two figures is definitional, and it maps the boundary of the rule with precision, showing exactly where protection currently applies and where extending it would add most.
Even after the additional costs carried by payment firms, estimated at £44m to £56m a year, the policy delivered a positive short-term net benefit of £17m to £29m. Frontier called that a conservative reading, since several of the benefits resist quantification altogether. Consumer protection measures that clear their own costs within a year are rare enough to change how similar proposals are assessed.
Outcomes vary to some degree depending on who a consumer banks with, and a roadmap published alongside the evaluation sets out how to bring every firm up to the standard the strongest performers have shown is achievable. A formal consultation follows in December 2026, giving the industry a direct hand in how consistency is delivered.
UK Finance recorded £1.28bn taken through payment fraud during 2025, of which domestic consumer APP fraud forms one portion. The rule has proved its effect on the losses it reaches, which makes the case for applying the same principle to the channels it does not yet touch a matter of arithmetic.
“The evidence is clear – APP reimbursement is working. Payment fraud losses are down, more victims are being reimbursed, and firms are investing in prevention. But we are not complacent. There is more to be done to ensure consistency in how consumers are treated, along with a step-change in the approach taken by tech firms and telcos to keep up with, if not outpace, those criminals exploiting their systems.”
David Geale, Managing Director, Payment Systems Regulator
The regulator’s own origin data points to where that effort would pay. Meta’s platforms were linked to 54% of scam incidents, while fraudulent calls and texts accounted for 12% of cases and 31.5% of losses. Fuller platform-level data is due before the end of the year.
The agentic dividend
The timing couldn’t be better, payment initiation by autonomous software agents is moving from demonstration into deployment, and the liability question underneath it remains open. When an agent is deceived, misinstructed or compromised and money leaves an account as a result, the loss has to land somewhere. Most of the frameworks drafted so far to answer that originate with American protocol designers and platform operators.
Britain brings something to the table. The evaluation supplies empirical support for a principle that had rested on theory alone: assign the loss to the party with the greatest capacity to prevent it, and total losses fall. Applied to agentic commerce, that points towards the operators of the agents and the platforms hosting them, working alongside the banks settling the payments. How regulators choose to allocate it is a separate decision, but the mechanism has now been tested at national scale and found to work.
A regulatory export
Legislating first produced the headlines, and evaluating first produces the influence. Standard-setting bodies and finance ministries weighing comparable rules have until now argued from projection; Britain can argue from a measured result, including on the objections that never materialised. That is a genuine asset in a debate about to be settled internationally.
Detection investment made under the current rules carries a documented return, which alters how such spending is argued for internally and how quickly it gets approved. Any organisation designing agentic payment services will find that the UK is where liability precedent is forming first, where the December consultation offers a seat at the table, and where the supporting evidence already exists in published form.