A financial institution declining or terminating relationships across a category rather than assessing each case, in response to perceived compliance cost. The category can be a sector, a jurisdiction, a payment type or a class of funding source, and the decision is taken once and applied to everyone inside it. Nobody in the category is accused of anything, and the file is not read and rejected — it is not read. The counterparty learns of it from an outcome — a returned transfer, a closed account, a relationship manager who cannot explain — rather than from a finding.
A cost decision, not a finding against the customer
The institution is answering a question about its own economics: what a category costs to hold, against what it earns. Assessing a counterparty individually consumes analyst time, escalation, senior sign-off and periodic review, and the cost falls due whether or not the relationship is retained. Exiting the category removes the cost permanently.
The asymmetry underneath it is what makes the outcome durable. Retaining a customer who later turns out to be a problem produces a supervisory finding, a penalty and a name in a press release. Refusing a customer who was never a problem produces nothing measurable at all. One error is visible and priced; the other is invisible and free. Any institution reading the two costs correctly refuses.
That is why a refusal carries no information about the counterparty, and why the usual responses to it — more documents, a longer explanation, a letter from an accountant — are answering a question the institution did not ask.
Where the decision is actually taken
The refusal that reaches a counterparty is frequently not made by the institution they are speaking to, and frequently not about them. Three distinct levels produce the same experience.
| Level | Who decides | What the counterparty is told | What evidence changes it |
|---|---|---|---|
| Customer exit | The institution holding the relationship | Notice of termination, normally without a reason | Little. The review has already closed |
| Product or segment withdrawal | A policy or risk function above the branch | That the product is no longer offered | Nothing at customer level — the product is gone |
| Correspondent withdrawal | A bank two institutions away, deciding about the respondent bank, not about its customers | Usually nothing. The payment fails, returns, or takes an unexplained additional week | Nothing the counterparty holds. The respondent has to replace the corridor |
The third level is the one that misleads. A correspondent withdrawing from a respondent bank is making a judgement about that bank’s jurisdiction, portfolio and supervision. Every customer of the respondent then loses the ability to send or receive in that currency without any of them being assessed, and without the respondent’s staff being able to say more than that the payment did not go through. The decision was already several institutions upstream when it reached the counterparty as a failed transfer.
The network effect of that level is measurable. The Bank for International Settlements’ Committee on Payments and Market Infrastructures, which tracks active correspondent relationships from payment-message data, recorded a contraction of roughly a quarter between 2011 and 2020, continuing across effectively all regions while payment volumes rose. Fewer institutions now carry more of the traffic, which concentrates the effect of each further withdrawal.
What changed in the standards between 2025 and 2026
The over-application of AML rules stopped being a commentary problem and entered the rule text.
In February 2025 the Financial Action Task Force amended Recommendation 1 and its interpretive note — the provision that carries the risk-based approach itself — with consequential amendments to the interpretive notes to Recommendations 10 and 15. The term “commensurate” was replaced by “proportionate” throughout the standards, and countries are now explicitly required to allow and encourage simplified measures where risk is lower. Revised guidance on AML/CFT measures and financial inclusion followed at the June 2025 plenary. Changing the standard rather than the guidance around it places the problem inside the rule.
In the European Union the direction is toward binding obligations on institutions. The European Banking Authority issued guidelines in March 2023 aimed at unwarranted de-risking and at safeguarding access to financial services. The 2024 AML Regulation, which applies from 10 July 2027, requires the Anti-Money Laundering Authority, jointly with the EBA, to issue guidelines by that same date on the measures credit and financial institutions may take when applying AML rules alongside the Payment Accounts Directive — expressly including the business relationships most affected by de-risking practices. AML/CFT mandates transferred from the EBA to AMLA on 1 January 2026, and AMLA takes direct supervision of forty selected financial institutions from 2028. The same regulation widens the perimeter: traders in precious metals and stones become obliged entities for transactions at or above €10,000, and commercial cash payments are capped at €10,000 across the Union, both from 10 July 2027.
In the United States the instrument is supervisory rather than prudential. An executive order signed in August 2025 directed the federal banking agencies to remove reputation risk from the material used to examine institutions. The Office of the Comptroller of the Currency and the Federal Deposit Insurance Corporation published a final rule codifying that removal on 10 April 2026, effective 9 June 2026; the Federal Reserve had ended the use of reputation risk in its examination programmes in June 2025 and proposed its own codifying rule in February 2026; and on 2 June 2026 the three agencies jointly stripped reputation-risk references from a set of interagency guidance documents.
The limit of that last change is the part that matters operationally. The rule binds the agencies. It prohibits a supervisor from criticising an institution on reputation-risk grounds, or pressing it to close an account; it does not require any institution to open or keep one, and it leaves commercial discretion untouched. A counterparty whose payment was refused on category grounds in 2026 holds the same position it held in 2024. The regulatory pressure to over-exclude has been removed; the cost asymmetry that produced the behaviour has not.
What it looks like when it happens to one payment
The sequence is invisible until it terminates. The payer’s institution releases the transfer. It reaches the receiving institution, where the review can run after the funds have arrived rather than before they were released. If the category rule bites, the payment is returned to the account it came from, on the returning institution’s timetable rather than the payer’s, and the counterparty is told only that it did not arrive.
The cost is a calendar, not a fee. It has four components, each of which runs independently of the amount:
- The dead interval. Funds are neither with the payer nor with the seller between release and return, and neither party can act on them.
- The currency movement. A payment released on one date and returned on another comes back after whatever the two currencies did in between; that difference sits with the payer.
- The expired quote. A fixed price has a stated validity window. A payment that returns after the window has closed does not resume the transaction — the transaction is re-quoted at the prevailing figure.
- The second review. A retry through another route arrives at an institution that has no record of the first attempt and every reason to ask why the payment is being presented a second time.
What goes wrong that is mistaken for de-risking
Four failures produce a refusal that looks categorical and is not, and each has a different remedy.
A payment that arrives without a reference the receiving institution can match to an expected transaction becomes an unattributed credit. It is held in suspense and generates enquiries — an outcome created by the payment’s construction, not by the payer’s category.
A payment released by an entity other than the account holder of record is a third-party payment, commonly refused on that ground alone. It is routinely reported back to the customer as a compliance refusal, which sends them off to gather evidence about a question nobody raised.
A file that has lapsed at periodic review will block a transaction in a way indistinguishable, from outside, from a category exit. This one is genuinely fixed by supplying a document, which is why the reflex to send documents survives.
And a single unexplained inbound payment, with no pattern behind it, can trigger a relationship review at an institution whose thresholds are set for retail behaviour rather than for the size of one bullion purchase.
De-risking, debanking, sanctions blocking and a suspicion-based exit
Four terms describe four different acts and are used interchangeably in practice.
De-risking is category-level and driven by cost. No individual assessment occurred.
Debanking carries a narrower statutory sense in the United States since 2025: denial or conditioning of service on the basis of political or religious belief, or of lawful but politically disfavoured activity. It is the belief-and-viewpoint version of the same outcome, and the 2026 rulemaking addresses that version specifically.
Sanctions blocking is a legal prohibition attaching to a named party. The funds are frozen rather than returned, the institution has no discretion, and no amount of documentation moves it.
A suspicion-based exit is the opposite of de-risking: one relationship assessed on its own facts and ended on them, sometimes following a report the institution cannot disclose.
The distinction is not academic. Only one of the four is answered by producing further evidence about the counterparty, and it is the one least often at issue.
What gets priced is the seller’s settlement path, not the buyer’s wealth
A buyer refused on category grounds while funding a purchase from digital assets almost always responds by documenting themselves: exchange records, holdings statements, an accountant’s letter, source of funds evidence in depth. That evidence is necessary for the seller’s own file and it changes nothing at the bank, because the receiving institution’s category rule does not look at the buyer’s wealth. It looks at the payment: what instrument arrived, from which regulated payer, against which reference, in which currency, for what stated purpose.
The buyer controls none of those four. The seller controls all of them, and set them before the buyer arrived.
Where the digital-asset leg closes at a licensed digital-asset platform — the platform screening the wallet and the transaction under its own licence, converting, and settling the seller in national currency — what subsequently reaches the banking system is a payment in national currency from a regulated payer, against a stated order reference, matched to a pro forma invoice for identified goods. The transaction the institution assesses is a commercial settlement between two identified parties. The question that categorical rules exist to answer is closed by the way the payment is built rather than argued after a refusal, and the digital-asset history is documentary evidence in the file rather than the instrument being presented.
This makes settlement architecture a property of the counterparty, testable in advance and in writing: who is the payer of record, what reference binds the payment to the order, what currency the seller receives, and whether the seller receives digital assets at all. A seller who cannot answer those four questions before an order is placed is a seller whose payments will continue to be assessed by category, and the buyer will absorb every failure as delay. The route itself is set out at Buy Gold with Crypto.
Golden Ark Reserve’s banking, settlement and payment-evidence arrangements are set out at Banking and Payments.