Banking intervention

Bank intervention matters because authorization is not the final opportunity to limit a scam loss. A bank may block or hold a payment before release. After funds are sent, financial institutions may attempt a recall, freeze onward movement, recover money or consider reimbursement. These actions protect the payment channel rather than asking a customer to determine whether a voice, video or message was synthetic.

The following diagram shows the intervention points documented in bank and recovery programs.¹⁴

flowchart TD
    A[Payment instruction] --> B{Funds already sent?}
    B -->|Not yet| C[Block or hold]
    B -->|Already sent| D[Recall or freeze]
    B -->|Already sent| E[Recovery review]
    B -->|Already sent| F[Reimbursement review]

Figure: Intervention can begin before release; after sending, recall, freezing, recovery and reimbursement review are separate possibilities.

Among customers of Australia’s four major banks during the nine months from July 2023 through March 2024, customers made A$941 million in scam transactions and 68,317 customers experienced financial loss. The banks detected and stopped 24% of scam transactions by value. Receiving financial institutions recovered 20% by value of scam funds that customers sent, while the banks reimbursed or compensated 6% of customer scam losses.¹ ASIC linked an earlier improvement in detection and stopping primarily to added friction for cryptocurrency-exchange payments, better holds and behavioural-biometric detection.¹

Policy can also change incentives. A July 2026 independent evaluation estimated that UK APP scam losses sent over Faster Payments fell after the October 2024 reimbursement requirement. Under that framework, sending and receiving payment providers share reimbursement liability equally, giving both sides a financial reason to prevent fraud.²³

A useful bank conversation therefore separates whether a transfer is pending, whether a recall or recovery attempt is possible, and whether a reimbursement policy applies. Each question concerns a different intervention point.

Boundary. The Australian rates cover specified banks and exclude attempted scams prevented before a customer performed a transaction.¹ They do not isolate synthetic voice, video or text. No cited study measures how often a particular warning, hold or staff call defeats a live AI-assisted impersonation. The UK evaluation found varying provider outcomes and left longer-term effects uncertain.³