How U.S. Banks Are Using AI to Fight Fraud—and Where Trust Falters

A new study highlights how banks in the U.S. are integrating AI into customer service, particularly for fraud resolution. The research shows varying levels of comfort among customers, depending on the tasks AI performs. Trust in banks remains high, but frustrations with communication persist. The report outlines a three-tier model for AI use in banking: Inform/Explain, Recommend, and Act. Customers feel most at ease with AI in explain-and-alert scenarios, such as transaction updates or fraud alerts. Comfort declines when AI recommends financial steps or takes direct actions like processing payments.

Escalation in fraud cases is triggered by repeated ambiguity, negative language, dispute signals, or high-value transactions. AI can initiate alerts, interpret responses, and suggest next steps, but human oversight is required for critical decisions. Clear guardrails are essential to ensure accuracy, smooth routing, and rapid escalation when needed.

Trust in banks to communicate updates clearly stands at 77%. However, 54% of customers reported at least one communication-related frustration in the past year. The study also found that 68% are comfortable with AI assisting in at least one aspect of their banking experience. AI roles in banking must align with customer trust levels, not just technical capabilities. Banks are advised to focus on well-defined guardrails and escalation protocols. This approach aims to balance efficiency with customer confidence in high-stakes interactions.