Visa on September 1, 2026, launched an enhanced version of A2A Protect, its fraud tool for account-to-account (A2A) payments. The main addition is a unified fraud score, which Visa calls its first in-market integration of technology from Featurespace, the fraud analytics company it acquired. The score is calculated in real time, before the funds leave the payer’s account. Visa says A2A Protect has increased fraud detection by 75% in the first six months of deployment.
The window to stop A2A fraud closes at the debit
Account-to-account transfers settle in central bank or commercial bank money, without the chargeback rights that come with cards. Once a transfer is executed, getting the money back depends on cooperation between banks, not on a right the payer holds. That makes the moment before the debit the one that counts. A2A Protect returns its signal before execution, while the payment can still be held.
One score replaces a patchwork of fraud signals
A bank fighting transfer fraud typically combines signals from several separate tools, each with its own scale and threshold. The new score rolls those signals into a single value, delivered through one API. Each alert also carries a plain-language explanation of why the transaction was flagged, written for the fraud teams who work the alerts.
- A single score, calculated at network level and returned in real time
- Integration with existing systems through a single API
- A plain-language reason for every flag
- Opt-in intelligence sharing across institutions to spot coordinated fraud and emerging scam hotspots
Transfer learning gives banks a head start, with a catch
The tool relies on transfer learning, which reuses a model trained on one data set for a related task. Applied to fraud, it lets a bank benefit from patterns seen elsewhere on the network without building up its own history first. Visa says banks don’t have to wait months for models to learn from their own transaction data, or for other banks to join a consortium. The catch is representativeness. A pattern common in one market can be rare in another, and the transfer is only as good as the resemblance between the two populations.
Visa names no clients, markets, or launch date
The announcement names no client bank and no market where the tool is live, and it gives no general availability date. The 75% gain in detection covers “the first six months of deployment,” with no baseline and no indication of which institutions were measured. A detection rate also has to be read alongside the false-positive rate, since the two tend to move together.
James Mirfin, Visa’s head of risk and security solutions, framed the product around speed: “Fraudsters move fast across payment types, and financial institutions need risk insights just as quickly, without slowing down legitimate payments.” That is the core constraint for a tool like this one. It has to return a verdict within the time it takes to execute an instant payment.