The Outlier Lab

Authorized Push Payment (APP) Scams

Hub note: scams in which the deceived customer authorizes and sends the payment themselves, covering typology, transaction footprint, intervention points and the shifting liability models in the US and UK.

What it is

APP scams invert the fraud model most stacks were built for. Every authentication control passes, because the device, the customer and the login behavior are all genuine. The anomaly is in the purpose of the payment and the customer's state of mind, not in who is at the keyboard. That's why APP is the central transaction-fraud problem for instant and irrevocable rails.

The Federal Reserve's FraudClassifier model first splits authorized from unauthorized by asking who initiated the payment. Its companion ScamClassifier model then asks whether the incident involved deception or manipulation for financial gain, and whether an authorized party was tricked into paying or into giving access. It then sorts scams into two categories (products or services; relationship and trust) and nine types:

ScamClassifier categoryTypes
Products and servicesMerchandise · Investment · Property sale or rental
Relationship and trustRomance impostor · Government impostor · Bank impostor · Business impostor · Relative/family/friend · Other trusted party

Adopting a shared taxonomy like this is the cheapest high-value step most programs can take. Without it, scam losses get labeled as "customer error," "ATO" or "dispute," and you can't measure them. See Loss Taxonomy and Attribution and Frameworks and Taxonomies.

Gray zones practitioners argue about:

  • Hybrid ATO and scam. A victim is talked into reading out an OTP or installing remote-access software, and the criminal then moves the money. Under most US rules this counts as unauthorized (Reg E), even though social engineering drove it. See MFA Bypass and OTP Interception and Regulation E and Unauthorized Transfers.
  • "Safe account" moves. The victim moves money into an account they believe is theirs or protected. This is classic APP. See Impersonation Scams.
  • Mule victims. Some recipients are themselves scam victims (romance, job or overpayment), which complicates both labeling and offboarding. See Money Mules.

How it works

Diagram: Lure or pretext (impersonation, romance, investment, purchase), Trust and urgency built, Coaching: what to say, which rail, why to ignore warnings, Victim initiates payment from own account, Mule or scammer account receives, Rapid layering (P2P, crypto, cash out)

All APP typologies share three properties that defenses can target:

  1. A pretext that explains why the payment is necessary and urgent.
  2. Coaching that neutralizes the bank's friction ("they'll ask why; say it's for a family member").
  3. A receiving account the criminal controls, usually a mule.

Transaction footprint

StageWhat you see in the dataDecision point
Pre-paymentInbound call or SMS contact patterns (if visible); unusual login timing; active phone call during the session; remote-access tools; hesitant navigationSession-level scam risk (Remote Access Tool Detection, Behavioral Biometrics)
Payee setupNew payee; payee name doesn't match account holder; payee seen across other customers' scam reportsPayee Confirmation and Payment Friction, Payee and Counterparty Risk
Funding the paymentMoving money between the customer's own accounts first (savings → checking), raising limits, liquidating investmentsTreat limit increases and liquidations as scam precursors
The paymentFirst-time high-value payment on an irrevocable rail; round amounts; purpose codes that don't fitReal-Time Transaction Risk Scoring; dynamic, contextual warnings (Customer Education and Scam Warnings)
RepetitionA sequence of payments to the same or new payees; rail switching after declinesCustomer-level velocity across rails
Receiving sideNew or dormant account receiving from many unrelated senders, then emptied quicklyFast In Fast Out Patterns; receiving-bank monitoring
ReportVictim reports days or weeks later, often after the scammer goes silentLabeling (ScamClassifier), recall attempts, reimbursement decision

Intervention works best before the first high-value payment to a new payee. Each later payment is harder to stop, because the victim is more committed and more heavily coached. Measure intervention effectiveness by first-payment catch rate as well as total dollars saved.

Common variants

Detailed in child notes:

VariantPretextTypical railNote
Bank, government and tech-support impersonation"Your account is compromised; move your money"Wire, P2P, cash, gold, kioskImpersonation Scams
Investment and pig butchering"Invest with me"Crypto, wire, ACHInvestment Scams (Pig Butchering)
RomanceRelationship, then emergencyWire, gift card, cryptoInvestment Scams (Pig Butchering), Elder Financial Exploitation
Purchase and marketplaceItem for sale that doesn't existP2P, instant paymentsSeller Non-Delivery and Fake Listings, Seller-Targeted Scams (Fake Payment and Overpayment)
Family emergency ("grandparent")Relative in trouble, often with a voice cloneCash courier, wire, gift cardGenerative AI and Deepfake Fraud, Elder Financial Exploitation
Gift-card paymentAny pretext; gift card as the railGift cardGift Card Fraud
Invoice and payment-redirection (consumer)Fake contractor or real-estate closing instructionsWireCorporate BEC is out of scope

Where it shows up

Impact

  • Scale (US): the FTC's 2025 Consumer Sentinel data recorded $15.9B in reported fraud losses. Imposter scams were the most-reported category (over 1M reports, over $3.5B), and investment scams were the largest by dollars ($7.9B). FTC data show bank transfers producing the highest aggregate losses, followed by cryptocurrency. IC3 recorded $20.9B in 2025 cybercrime losses overall, 85% of it cyber-enabled fraud.
  • Liability: in the US, authorized payments generally fall outside Reg E's unauthorized-transfer protection, so the loss lands on the victim unless the institution or network chooses otherwise. That baseline is under pressure (see Case Study - Zelle Scam Liability Disputes (2024-2025)). The UK moved to mandatory reimbursement in October 2024 (see Case Study - UK APP Scam Reimbursement Rules (2024)).
  • UK reimbursement model: since 7 October 2024, sending PSPs must reimburse in-scope APP scam victims on Faster Payments (under PSR direction) and on CHAPS (under the Bank of England). The main parameters:
    • Cap and excess. Reimbursement is capped at £85,000 per claim. The PSR had set £415,000 in December 2023, then cut it in October 2024. Senders may apply an excess of up to £100.
    • Cost split. The receiving PSP repays the sender half of each reimbursement.
    • Exceptions and window. Claims can be refused for first-party fraud or gross negligence (the "consumer standard of caution"), and when made more than 13 months after the final scam payment. Neither the caution exception nor the excess applies to vulnerable customers.
    • Timing. Claims are paid within five business days unless the sender stops the clock to investigate, and must be closed within 35 business days.
  • UK results. The PSR's dashboard reports that 88% of money lost in in-scope claims (£316M) was reimbursed from October 2024 to March 2026. Only about 3% of claims were rejected under the caution exception. A Frontier Economics evaluation the PSR published in July 2026 estimated:
    • Faster Payments APP losses fell by about £73M a year, and scam cases by nearly 35,000.
    • Reimbursement across all claims, including out-of-scope ones, rose only from 54% to 65%, while 97% of in-scope claims are now reimbursed. Scope boundaries matter as much as the headline rate.
    • The PSR plans a consultation before the end of 2026 to make firms apply the rules more consistently.
  • UK regulator change. HM Treasury confirmed in April 2026 that the PSR will be abolished and its functions folded into the FCA. That requires primary legislation, and no timetable had been set. Transitional legislation is to preserve existing PSR requirements, including the reimbursement regime.
  • Receiving-side exposure: mule accounts carry BSA/AML exposure regardless of who reimburses (BSA-AML and SAR Filing).

Red flags

Prevention and detection at a glance

LayerControl or signalWhat it does here
PreventPayee Confirmation and Payment Friction (name match, cooling-off on new payees)Breaks misdirection and buys time
PreventCustomer Education and Scam Warnings, specific to the paymentGeneric warnings get clicked through; typology-specific questions work better
PreventVelocity and Exposure Limits on new payees and new railsCaps first-payment exposure
DetectReal-Time Transaction Risk Scoring with session and customer contextHas to score intent, since authentication will pass
DetectReceiving-side Transaction Monitoring and Graph AnalyticsMules are the shared bottleneck across typologies
RespondHuman intervention calls from dedicated Manual Review Queues; Trusted Contact and Senior Investor HoldsTrained specialists outperform scripts for coached victims
RecoverFBI IC3 Recovery Asset Team (domestic wires), FinCEN Rapid Response Program (international), Consortium and Data SharingIC3 reports a 58% freeze success rate on 2025 kill-chain referrals; speed matters

Emerging variants and adversary adaptation

  • Escalating multi-persona scripts. IC3's November 2025 PSA describes criminals posing as bank, payroll or tech-support staff, then handing off to a second persona posing as law enforcement. The PSA counted more than 5,100 account-takeover complaints from January 2025 onward, with losses over $262M. Scam pretexts increasingly end in the criminal moving the money, which blurs authorized and unauthorized.
  • Offline cash-outs to get around digital friction. Courier pickups of cash and gold have grown. IC3 counted about 725 "gold courier" complaints totaling $311.8M in 2025, and a 2026 PSA describes couriers stepping in specifically when bank transfers are blocked.
  • AI in the pretext. Voice clones for family-emergency scams, deepfaked public figures for investment lures, and AI-generated scripts. IC3 counted 22,364 complaints with an AI nexus in 2025 (about $893M). See Generative AI and Deepfake Fraud.
  • Liability shift drives adaptation. Where receiving institutions share liability, as in the UK, the economics move toward stopping mule accounts. Expect criminals to favor rails and jurisdictions outside reimbursement scope (crypto, cash, international transfers).
  • Better data sharing. FinCEN's June 2026 314(b) fact sheet confirms that real-time sharing, attempted transactions, and mule-recruitment information are all in scope. That opens the door to sender-receiver bank collaboration on live scams.

Real-world examples

Sources and further reading