A research team at the University of Massachusetts Amherst has found three types of problems in ACH, the US interbank payment network, in which bank balances or credit card limits become usable before the debit of the underlying funds has succeeded. The team modeled the payment rules to identify the problems, then reproduced the phenomena using real bank accounts that the researchers themselves controlled. The study covered two banks, three card issuers and three payment platforms. According to the authors, all of the reported problems had been fixed by January 2026. In an explainer published by the research team, they describe the core issue as a time gap between funds becoming usable in an account and the ability to reliably collect the underlying funds from another bank.

AD

The Gap Between Available Balances and Interbank Settlement

In ACH, authorizing a payment, posting it to each financial institution's internal books, and settling between banks proceed as separate processes. Even when the receiving institution makes a provisional deposit available, the corresponding debit at the sending bank may not yet be final. If that bank then rejects the debit request and reverses the transaction, an event known as a "return," the funds that were supposed to back the balance or credit limit that has already been made available never arrive.

The team named this class of problem "Credit-Without-Debit." On the phenomena observed in the field tests, the research page explains that the existence of an available balance or credit limit did not guarantee that the underlying debit would succeed. The researchers point out that completing processing at an individual institution must be distinguished from the transfer as a whole completing successfully.

For example, being able to deposit funds into a bank account does not mean funds can be debited from it. If a debit block is set on the account, a deposit may be accepted but the later collection can be rejected. Likewise, even if a card company records a payment internally and restores the credit limit, the debit from the paying bank may fail afterward. The problem lies in treating success at an intermediate stage as grounds to assume that the subsequent debit will succeed too.

From Modeling ACH Rules to Testing Real Accounts

The research was conducted by Ali Hamza Malik and Muhammad Taqi Raza of the University of Massachusetts Amherst. Their paper, "Credit Without Debit: Formal Verification of Banking Payment Systems," was peer-reviewed and accepted to ACM CCS 2026, an international computer security conference. It appears in the conference proceedings under DOI 10.1145/3830454.3846614. The authors and title can be confirmed on the official list of accepted papers for ACM CCS 2026, and the call for papers describes the review process.

Based on Nacha's 728-page ACH rules, the team built a model centered on "WEB," domestic consumer online payments, and "PPD," payments based on prior authorization such as recurring billing. They represented the processing of each party involved in a payment as five mechanisms that communicate while changing state, connected by two communication channels, and analyzed the model with the model-checking tool nuXmv. Model checking is a technique that systematically examines the possible sequences of processing in a model and finds, as "counterexamples," specific sequences that violate defined safety conditions.

They verified three safety requirements. First, debit authorization must remain valid from transmission through settlement. Second, for both small account-verification deposits and ordinary payments, debits and credits must balance once the full sequence of processing is complete. Third, funds or credit must not be made usable until after the period in which a transaction can be returned has passed. According to the description of the published verification software, counterexamples were found for all three requirements. However, these results were obtained on a model limited to a finite scope and do not indicate the number or amount of losses that have actually occurred at real banks.

The team then tested real financial services using consumer accounts they owned. The targets were two of the four major US banks handling ACH transactions, three card issuers and three payment platforms. What they checked were items visible on customer-facing screens, such as balances, payment statuses, processing times and return codes; they did not directly examine financial institutions' internal logs. At the institutions' request, names were anonymized, and the banks are referred to as "Bank A" and "Bank B." The experiments reproduced problems arising from the order of payment processing in a limited environment, and are not a study of how often such problems occur among ordinary users or of actual crime damage.

AD

Three Problems: Observed Phenomena and Conditions for Loss

The problems the team confirmed fall into three types: early use of a provisional deposit, failure to recover a small account-verification deposit, and early restoration of credit card limits. The nature of the funds or credit that became usable differs for each, and whether an actual loss occurred must be distinguished as well.

Problem type State that became usable first Result on the funding side Observed outcome and conditions for loss
creditBLEED Bank A made a provisional deposit available on day 3 of the transaction About 2 days later, Bank B returned it with R16 In the test, Bank A reversed the deposit. If funds are withdrawn before the return, a loss results
creditMINT Three platforms made small verification deposits on days 1–3 Collection on days 4–5 failed with R08 or R16 Deposits of $0.01–$0.99 each were left uncollected. The authors note the deposits could be repeated
creditRESET Three card issuers fully restored credit limits within 2 business days The funding debit failed about 6 business days later The restored limit was usable for about 4 days. New payments made in that time would result in a loss

Note: Based on Figure 3 and the field-test descriptions on the research team's published page. R08 and R16 are ACH return codes. The timeline in the figure uses approximate business days, and the text also mentions "3 days later" for creditBLEED.

The roughly two days observed in creditBLEED is the period during which the provisional deposit could be used without the funding debit being final. In the test conducted under the researchers' control, Bank A ultimately reversed the deposit, so funds were not actually taken out and no loss occurred. In creditMINT, by contrast, the process of recovering the small funds sent for account verification itself failed. What was confirmed in creditRESET was the time gap between the card company restoring the credit limit and the funding debit failing.

Therefore, simply comparing creditBLEED's roughly two days with creditRESET's roughly four days does not tell us which problem could lead to greater financial harm. Funds or credit becoming usable, the funding being uncollectable, and funds actually being spent and causing a loss are distinct stages. The research team also did not estimate total damage across ACH as a whole.

Fixed by January 2026, but Design Questions Remain

The team reported the problems and proposed countermeasures through the relevant organizations' official channels on April 16, 2025, and later supplied additional evidence. According to the authors, creditBLEED and creditRESET could no longer be reproduced by May of that year. For creditMINT, the problem persisted at one company until December of that year, but was resolved at all targeted providers by January 2026, which the team confirmed through follow-up testing. However, the team has not received technical explanations of the fixes each company made to its internal systems.

The countermeasures the authors propose are a design that rechecks debit authorization before payment and restricts the use of provisional deposits until the funds have passed the access controls of the bank supplying them. They also call for confirming, when linking bank accounts, not only that the account can receive deposits but that debits can be made from it. They further recommend treating a failure to recover a small verification deposit as a failure of the account verification itself, and revising mechanisms so that retries do not simply repeat the deposit. These are countermeasures proposed by the research team and should be distinguished from the fixes each company actually adopted.

The models, safety requirements and scripts for reproducing the counterexamples used in the verification have been made public. Version 1, archived on Zenodo, has been assigned the DOI 10.5281/zenodo.22701199. The screens and transaction logs obtained in the field tests, on the other hand, have not been released because they contain personal information related to financial transactions. The number of individual accounts used in the tests and the total number of trials are also not given, and no successful independent replication by a third party has been confirmed.

The model mainly targets domestic US consumer ACH payments and does not include wire transfers, real-time payments (RTP) or the credit card payment networks themselves. Although the three problems studied have been fixed, the design question remains of when funds or credit should be made usable in payments across different financial institutions. What matters is how far the funding-side debit has been confirmed before the receiving side allows the funds to be used. Connecting the processing that proceeds at individual institutions to reliable completion of the transfer as a whole will require those confirmation conditions to be made clear.