Beyond the victims and the billions of dollars in losses, the rise of AI-driven fraud means banks, FinTechs and payment providers must think about risk, liability and real-time fund flows.
Fraudsters are using the same intelligence tools that power modern commerce to increase the speed, scale and success rate of attacks. The result is a widening gap between how fast fraud evolves and how slowly traditional refund and liability frameworks respond.
Refund and Return Rules Are Not Built for AI Fraud
This faster and more adaptive threat environment has exposed a growing mismatch between traditional refund eligibility rules and the realities of AI-driven fraud. Banks historically adjudicated refunds based on static criteria tied to transaction type, customer history or manual review.
But now, with fraud now shifting toward rapid credential compromise and unauthorized use, those frameworks are being strained.
Institutions report that fraud is no longer just a financial loss issue, as detailed in the PYMNTS Intelligence “2025 State of Fraud and Financial Crime in the United States” report commissioned by Block.
According to report, 50% of financial institutions say fraud erodes customer loyalty, 44% report brand and reputation damage and 48% cite lost business opportunities.
Advertisement: Scroll to Continue
These trust-based consequences intensify pressure to make refund decisions quickly and consistently, especially when consumers expect instant resolution against a backdrop where funds can move instantly between accounts.
AI-Driven Scams Accelerate Across Channels
Fraudsters are now weaponizing artificial intelligence across scams that hinge on deepfakes, impersonation and the exploitation of personal data. These attacks feed directly into the surge of unauthorized-party fraud, which now represents 71% of all fraud incidents and dollar losses, reversing last year’s pattern in which authorized-party manipulation was more common.
The Block and PYMNTS Intelligence data show credential compromise, account takeovers and impersonation scams climbing sharply, with digital-payment fraud accounting for as much as 20.3% of total fraud losses on a dollar-weighted basis.
These patterns show how AI amplifies the ability to mimic legitimate identity signals or automate large batches of attacks that bypass static controls.
Instant Money Movement Makes Liability a Real-Time Issue
The proliferation of instant payments has made these tensions even more acute. Real-time fund flows mean institutions must “score” the transaction before it settles, not (sometimes hours) later, as had been the norm.
The operational pressure and challenge to so is clear in the data. The report found 46% of institutions say faster payments are a top fraud-management challenge, and 41% cite the expansion of payment types and currencies, including peer-to-peer and instant transfers.
This places liability questions squarely inside the transaction flow, where institutions must authenticate the actor, verify the action and determine whether the customer would be eligible for reimbursement if the transaction is later deemed fraudulent.
AI and Behavioral Analytics Provide a Line of Defense
To keep pace, financial institutions are rapidly modernizing fraud defenses. The PYMNTS Intelligence/Block report shows that behavioral analytics is now used by 70% of institutions, while machine learning is used by 61%.
These systems examine behavioral signals that are difficult for fraudsters to spoof, even with AI tools. Device fingerprinting, velocity checks, transactional context and behavioral baselines help determine whether a transaction aligns with the legitimate customer’s historical patterns.
The report also shows how embedded AI is transforming decisioning. Seventy percent of institutions say machine learning helps them balance proactive and reactive strategies simultaneously, and 25% say it makes their defenses predominantly more proactive.
This reflects the shift toward real-time detection across instant rails, where the goal is to intercept fraudulent attempts before funds leave the account.
These moves indicate recognition that manual and rules-based interventions cannot support high-velocity refund and liability determinations for instant transactions.
Expectations around fraud resolution are changing quickly. Faster payments compress the time for institutions to detect, investigate and decide whether a customer should be reimbursed. Artificial intelligence and behavioral analytics address this by sharpening identity assurance and enabling earlier signals of abnormal activity.
Institutions that modernize defenses not only reduce fraud losses but also strengthen customer confidence. The path forward is clear: real-time payments require real-time fraud intelligence, and the institutions that integrate AI deeply into risk scoring, identity proofing and refund governance will be better positioned to contain liability and maintain trust.
