The Bangkok Blueprint: protecting payments that victims authorise

A policy framework, rather than a new compensation law

The International Monetary Fund (IMF) and World Bank Group (WBG) released the Bangkok Blueprint on 14 October 2026: 12 considerations, five pillars, Bank of Thailand coordination. The International Monetary and Financial Committee (IMFC) welcomed it on 17 October.^[1][2]

Its economic significance is that safety can be examined through the design and burdens of providing financial services, beyond asking users to be careful. An application can be technically secure while a customer, believing a deceptive story, instructs a payment that the system processes normally. A successfully executed transfer and a successfully completed purchase or investment are different outcomes.

Domestic law, contracts and the scope of compensation still determine individual rights. Publication of the framework does not establish a worldwide right to a full bank refund. The changes that matter will be the prevention measures and allocation of losses adopted in each jurisdiction. Those implementation choices connect a joint policy announcement to the claims an individual customer can actually make.

What authentication can establish

Consider a hypothetical impostor who claims to represent a real supplier and announces a change of bank details. An employee may log in on their own device and complete authentication correctly while acting on false information. The bank sees a different pattern from an account takeover using stolen credentials. Establishing that the genuine customer acted becomes less useful as a complete test of safety.

Three questions help: who gave the instruction, who will receive the money, and what the payment is intended to achieve. The first can be answered correctly while misunderstandings remain in the other two. Even a matching payee name does not prove that the recipient will deliver the promised goods. Authentication and payee checks remain valuable; understanding their limits makes the gaps more specific.

This is a design example, not an account of a transfer approved by a particular bank. It shows what a safety test based entirely on successful authentication can miss. Banks also face limits on examining every payment’s purpose, and excessive scrutiny of ordinary life and commerce would carry its own costs. The placement of additional checks affects both safety and usability.

European evidence: rates, amounts and who bears losses

The European Banking Authority (EBA) and European Central Bank (ECB) report that, in the European Economic Area (EEA) in 2024, fraud was about 0.002% of payment value; users bore roughly 85% of credit-transfer fraud losses.^[3]

A low value-based fraud rate and a large loss can coexist. With a very large payment base, a small fraction can still be devastating to those affected. Conversely, a large aggregate loss does not show that most digital payments are unsafe. Evaluating the cost of withdrawing from a service requires separating fraud value as a share of payment value from the burden when harm occurs.

The chart provides background predating the joint framework, not a measure of its results. Even with better security, remaining losses concentrated on a few users can still impose severe burdens. Keeping the aggregate value-based fraud rate low and protecting a household or business after an individual loss are both relevant objectives. Treating the first as proof of the second overlooks the role of compensation and accessible support.

SG Group View: measure harm and its allocation together

A single monetary measure can conflate different improvements. Preventing a fraudulent transfer, compensating a customer after funds have left, and recovering money from criminals change different flows. Reducing the victim’s burden is valuable in its own right, but it does not establish an equivalent reduction in criminal proceeds.

SG Group would examine the loss left with the victim, the burden shifted to the compensating party, and the funds recovered from criminals separately. These are not three amounts to add indiscriminately. They can overlap when recovered funds finance compensation, while some losses may fall outside a claims process. Defining the population prevents a reimbursement improvement from being presented as a prevention result.

This approach gives both protection and deterrence their proper weight. A system could return money promptly to a customer while recovery and investigation continue on a different timetable. Waiting for recovery before compensating the victim may deliver the same eventual amount but impose greater disruption to a household or business in the meantime. Duration matters alongside the identity of the party bearing the loss.

Reimbursement changes the incentive to prevent fraud

If a payment provider bears little of a fraud loss, its cost of making extra checks can be separated from the benefit those checks deliver to customers. A reimbursement obligation gives prevention a more direct financial value to the provider. Yet the party carrying the cost may lack some of the information needed to prevent harm. The initial deception can occur on another service, before an instruction reaches the bank.

Consider a design that also places costs on receiving-side providers. It can give firms with no direct relationship to victims a reason to prevent misuse. Choosing the allocation requires examining the information each party holds and the actions it can take to reduce losses. A fixed split simplifies administration but may not represent individual responsibility precisely. This is an analytical design choice, not a worldwide rule established by the joint release.

What the UK reimbursement data can show

The UK Payment Systems Regulator (PSR) reports closed UK domestic Faster Payments claims for Q1 2026: £72.6m reimbursed. Its figures distinguish reimbursable value from claims resolved, including rejections. Sending-firm information was not validated against receiving firms.^[4]

A high value-based rate does not mean every individual claim receives a full refund. Returning one large claim can move the ratio, while rejecting many small claims may have a limited effect on value. Fast resolution can likewise mean either prompt reimbursement or prompt rejection. Understanding personal protection requires matching aggregate results to the conditions defining an eligible claim.

The evidence nevertheless permits concrete scrutiny of operation. Separating money reimbursed, claims decided and time taken allows different questions about amount and speed. Following consistent definitions, and reading the reasons for exclusion, is more informative than assigning the entire difference from a differently scoped earlier survey to the policy. This makes both the strengths and the limits easier to identify.

The waiting period affects households differently

Two people receiving the same reimbursement can experience very different disruption while waiting. Someone with rent and living expenses due soon has a different exposure from someone with substantial reserves. An explanation that reimbursement is likely does not provide spendable money before a decision is reached and funds actually return.

This difference should not be confused with the degree of a victim’s carelessness. Limited cash makes an identical loss harder to absorb; it does not determine responsibility for the deception. Protection can be evaluated through the period without essential funds, the effort needed to find help and the treatment of hardship during a claim, alongside the final reimbursement percentage.

The guide to monthly household surpluses and immediately available cash explains this distinction. Holding reserves does not remove the need for fraud protection. Rather, measuring a victim’s burden solely by its eventual monetary value overlooks the difficulty before money returns. Providing timely reimbursement while identifying false claims is part of the design problem.

For businesses, payee changes test internal approval

Business payments have internal approval processes as well as bank authentication. Whether one person handles an order, invoice and account-detail change, or another person checks them, affects where a suspicious change might be detected. An invoice can match the order amount while only the receiving account has been altered. A correct amount does not validate the whole transaction.

Checks also impose different costs on new suppliers, recurring payments and unexpected account changes. Delaying every payment identically can disrupt legitimate suppliers. Conversely, routinely waiving checks on urgent transactions rewards a fraudster’s pressure to hurry. A workable process creates a place to examine changes without depending entirely on an overloaded employee’s vigilance.

Payment to the wrong account may leave an obligation to the real supplier outstanding, depending on the contract and law. A bank’s completion message alone does not establish shipment or discharge of the debt. Understanding how supply chains connect delivery and cash clarifies how a payment error could reach procurement. The payment and the physical transaction need separate examination, without presuming liability under an individual contract.

Mistaken blocks have costs too

If the sole objective were to let no fraudulent payment through, stopping every transfer would be an extreme solution. It would also eliminate the service’s purpose. In practice, controls can allow fraud or mistakenly interrupt genuine transactions. Neither error can be assumed away when calculating the benefits of additional checks.

A mistaken block requires inquiries, resubmission and explanations to the recipient. Missing a recipient’s deadline may affect the provision of goods or services. Fewer blocked transactions could reflect better decisions or weaker prevention. Fraud prevented and legitimate payments completed need examination within comparable groups of transactions.

Whether a suspicious instruction should be held, questioned or refused depends on circumstances and the governing regime. Support for the Blueprint does not itself grant banks a new power to stop payments. A useful comparison considers review routes and the burden of restoring a wrongly blocked user’s normal access, alongside the intensity of intervention. A trusted service must remain usable by legitimate customers.

More checks require capacity to resolve them

It is easy to add another question to a customer’s screen. The harder task is deciding whether an answer resolves the concern or requires investigation. Repeated warnings followed by repeated approval may add no information. Treating the number of questions as the strength of protection risks creating customer fatigue and a backlog of unresolved cases.

Human review is useful only within the reviewer’s actual capability. A reviewer may ask about a payment’s background without being able to investigate an entire supplier’s business. Matching signals to evidence that can resolve uncertainty helps focus inquiries. Concentrating every decision in a few experienced staff can turn their absence or workload into a new access constraint.

Our earlier article on G7 AI tabletop exercises and official reporting examines the exchange of information during an incident. Here the question is closer to the customer: can someone who has answered a check obtain a decision allowing or refusing the payment? International coordination does not automatically deliver that customer service. Staffing and authority to revise decisions belong within the service design.

Shared responsibility can mean different regimes

The Monetary Authority of Singapore (MAS) describes its Shared Responsibility Framework with the Infocomm Media Development Authority (IMDA) as covering a defined scope of phishing scams. It assigns duties to financial institutions and telecoms, with expectations of payouts when those duties are breached. It took effect on 16 December 2024.^[5] This is not the same regime or scope as UK APP reimbursement.

International comparisons can mislead if they begin with refund percentages. Eligibility, whether customer-instructed transactions are covered, and the duty breach that triggers a decision change the meaning of compensation. Rather than ranking regimes with different denominators, tracing which aspects of a loss reach which remedy better reveals practical protection.

The Blueprint’s international role is different from immediately unifying domestic systems. It may provide common ground for examining gaps, but the authority and resources to close them require concrete choices. For households and businesses using foreign services, the contracting entity and applicable regime are a starting point for understanding claims, beyond a familiar brand name.

Frozen balances and funds available to victims

A successful freeze can leave return procedures outstanding. Questions about ownership and competing claims can affect the decisions required before return. Counting every frozen balance as recovered money already available to victims would exaggerate results. Published amounts need to identify the procedural stage reached.

When a provider discloses compensation costs, cash payment and expense recognition may also occur at different times. Connecting expenses and cash through the financial statements helps interpret the customer outcome and the provider’s accounts. Higher reimbursement costs alone establish neither exhausted bank liquidity nor failed prevention. More claims, wider protection and higher underlying fraud are different explanations requiring different evidence.

Users bear more than explicit fees

The displayed transfer fee does not capture all the cost of protection. Repeating identity checks, contacting support after a block and travelling to a branch also consume resources. Small repeated checks can accumulate for frequent low-value payments. The same amount of time has a different proportional weight when applied once to a large transaction.

That does not justify simply dropping checks for small payments. Many small losses can form a fraud business, and even a small loss can remove money essential to a household. Combining transaction size and user impact into one measure risks leaving gaps. Costs and benefits should be assessed against the payment’s purpose and the alternatives available to its user.

Providers could recover costs through prices, limits, screening or changes to features. These are possible channels, not changes already announced as a result of the Blueprint. Any subsequent change needs examination of who it covers and which services it affects. A bank paying reimbursement does not mean its shareholders necessarily bear the entire eventual cost; competition and contracts affect where it lands.

Can smaller providers sustain the same protection?

A large bank may find it easier to maintain specialised fraud investigation and customer-support teams. For a smaller provider, development and overnight staffing can impose a higher fixed cost per transaction. Shared infrastructure or external services may reduce that burden, without eliminating the need to judge outsourced work and answer customers.

The competitive effect can run in either direction. Reliable baseline protection may help a small provider earn trust and win customers. Excessive fixed costs may instead discourage entry or small-value services. An identical requirement need not create an identical burden. Customer groups and payment characteristics matter alongside size.

This argument also does not establish that weaker protection improves competition. A cheap service that leaves losses with users or other providers undermines a fair price comparison. The question is which costs can be shared, which decisions remain firm-specific, and how comparable protection can coexist with different implementations. Separating the protection floor from implementation choices offers a route to both safety and choice.

Between the place of persuasion and the payment

When persuasion and payment occur in different places, improving one company’s interface has limits. Product promotion, conversation and transfer instructions may span several providers, with no participant seeing the whole sequence. A bank may observe the transfer without knowing the promise made to its customer; a platform may not know the ultimate receiving account.

Collecting everything is not automatically preferable. Conversations and transaction records can contain private or commercially sensitive information unrelated to fraud. Unfocused sharing increases the potential cost of a leak and the spread of mistaken assessments. Useful design identifies the information needed for a decision and the responsibilities of those permitted to use it.

The distribution of burdens also affects the incentives of nonfinancial businesses. If a bank pays compensation while a business earning revenue from the original promotion bears none of the cost, prevention benefits and expenses can diverge. Conversely, a uniform allocation of every loss merely because a company provided advertising or communications requires justification. Asking who knew about a risk and could change the outcome makes shared responsibility a more concrete choice.

Protection can change across borders

A payment that appears unified to a user can involve different contracting entities, transfer providers and account locations. A domestic compensation regime does not establish identical protection for international payments. Finding a complaints desk also does not prove it has power to stop a foreign account or order another country’s provider to reimburse a loss.

The economic consequence is that a cross-border service’s displayed price can omit the burden of obtaining redress. Additional languages, documents and contacts raise the effort required after harm. This is not an argument that all international payments are unsafe. It means that providers offering similar functionality can have materially different claim routes and protection.

In another field, the FSB stablecoin report’s distinction between holders and redemption entities illustrates why a service name is not the same as a counterparty to a claim. Stablecoin redemption and fraud compensation are different rights, with different obligations. Changing technology does not by itself resolve redress: the relevant entity and claim must be examined within each arrangement.

The economic effects go beyond adding losses

Money lost to fraud can affect household consumption or payments; for a business, it can remove funds intended for procurement or wages. Reported losses cannot simply be relabelled as lost world output. A transfer of money differs from a reduction in production, and reimbursement, recovery and access to replacement funding alter the outcome.

A broader channel runs through trust and use. If people retreat from digital payments to cash or face-to-face procedures, travel, storage and processing costs may change. Small firms serving distant customers could be particularly exposed. Conversely, confidence in protection may encourage previously excluded users to participate. These are conditional transmission channels, not observed changes following this announcement.

Effects on banks and payment firms also extend beyond compensation costs. Security spending, customer service, volumes and price competition affect profitability. Investors gain more from tracking persistent costs and changes in service use than inferring a stock-price outcome from a regulatory headline. Separating revenue, profit and cash helps interpret a business in which transaction growth and reimbursement burdens rise together.

More reported harm can have several explanations

An increase in reports can reflect both more fraud and easier reporting. People who previously gave up may begin to seek help, raising recorded cases. A decline can similarly mean less harm or greater difficulty reaching a complaints process. When eligibility changes, the population being compared changes too.

Evaluation should therefore separate counts, amounts, eligibility and time to decision. A few additional large cases can lift value while counts remain flat. Broader redress for small losses can reduce the average amount. Changes in an average need examination of which losses and users have entered the data, rather than a direct verdict on protection quality.

Comparable observation periods are essential. Losses occurring during a period need not be the same as claims resolved during it. Clearing a stock of difficult older claims can change a quarter’s results. Identifying whether a publication uses occurrence, reporting or closure dates helps interpret early implementation figures and later revisions without overstatement.

Testing the concern that reimbursement weakens caution

Does reimbursement make users less careful? The objection combines the benefit of placing costs on parties able to prevent harm with the possibility that users change their behaviour. Having a loss covered does not remove the distress of fraud or the time spent claiming. A regime must also address abuse of the claims process.

Testing the objection requires examining changes in user behaviour and transaction characteristics around implementation. More reported harm alone does not establish weaker caution: expectations of redress can encourage reporting. Rejections based on customer care also depend on decision criteria and claimant composition. Evidence of a behavioural side effect needs to be separated from evidence that access to the scheme has widened.

There are many choices between returning all responsibility to victims and accepting every claim unconditionally. Warnings must be understandable, individual circumstances considered, and evidence used to identify false claims. Capacity to investigate abuse while preserving timely redress would answer part of the objection. Suspicions that routinely obstruct genuine claims would instead separate the promise of protection from the customer’s experience.

Three paths, distinguished by observable changes

The favourable path combines better pre-payment decisions, usable legitimate transfers and accessible redress. Evidence would include reduced fraud within a consistent scope, lower burdens from mistaken blocks and appropriate resolution of eligible claims. An improvement in one dimension needs checking for costs shifted into the others.

A second path delivers better reimbursement without reducing criminal proceeds, leaving providers with a growing burden. Customer protection retains value, but prices or access conditions may change. A third path makes legitimate services harder to use and pushes activity into less protected channels. Better numbers at one firm do not establish less harm overall.

There is no basis for assigning fixed probabilities, and different payment types within one country could follow different paths. Building scenarios from explicit conditions is useful for recording which evidence supports which path, rather than ranking forecasts. When consistently defined observations arrive, the reasons for changing an assessment can then be explained.

Domestic implementation remains the open question

The joint release does not determine which domestic rules will change, when they will apply or how much each party will pay. Eligibility, decision deadlines and allocations between providers cannot be filled in from the framework’s five-pillar structure. International support does not make implementation uniform, and assuming it does would hide meaningful differences for users.

The next evidence consists of specific national proposals and decisions, provider terms, and consistently defined data on harm and redress. Commencement depends on specific domestic decisions. A proposal, final decision, existing arrangement and future change have different implications for users. Only a defined regime permits a proper comparison of scope and timing.

The approach in the investment-thesis and disconfirming-evidence template helps maintain a record. A view that fraud can fall while mistaken blocks decline requires evidence of both. Lower reported harm accompanied by lost users, or a narrower claims definition alone, would change that assessment. This is a way to discipline interpretation of new evidence, rather than a trading instruction.

The test is finance that people can safely keep using

The Blueprint’s value lies less in forceful international language than in directing attention to the protection needed for digital finance to grow. Authentication, detection, support for genuine use and post-loss allocation interact. Treating security as another department’s concern while counting only usage growth obscures the costs between them.

Shared responsibility can also diffuse final decision-making. A process is difficult to use if a victim must first determine whether the bank, telecom or platform is at fault before obtaining help. Internal allocation and the public route to redress can be designed separately. A clear claim route that does not push coordination disputes back onto the customer would make cooperation tangible.

The ultimate assessment should examine reduced victim burdens, prevention incentives reaching relevant participants and continued access for legitimate users. Higher reimbursement, more checks and international endorsement are evidence at different stages. Following changes in people’s funds and choices connects safety spending to its practical purpose: supporting everyday life and commerce.