AI hallucinations are a significant frontline CX and governance risk for financial services, threatening to create customer harm, compliance exposure, and reputational risk.
In fact, Glia’s benchmark report reveals that global losses from AI hallucinations reached $67.4BN in 2024, underscoring that hallucinations are no longer a theoretical technical flaw, but a material business and CX risk.
Financial institutions that move early toward hallucination-free, controlled AI will be better positioned to scale automation without sacrificing trust, setting a higher standard for reliability while reducing exposure across CX, compliance, and reputation.
Justin DiPietro, Chief Strategy Officer & Co-Founder at Glia, argues that because banking customers rely on information for real financial decisions, AI must be completely accurate every time.
“In banking, whenever there’s payments, transactions, whenever people are making life decisions based on the information that their bank is giving, there’s not an option to be wrong,” he explained.
“You can’t be probabilistically correct. You have to be 100% correct.”
Why “99.99% Accurate” Still Breaks the Bank
Hallucinations in banking create a much significant level of risk than error in ordinary retail settings.
Whilst a wrong product recommendation in retail can frustrate a shopper, an incorrect instruction during a funds transfer can cause direct financial lost.
“At every 10,000 orders for dog food, I get cat food, not that big of a deal. But at every 10,000 money transfers, I transfer the wrong amount, that’s an issue. That can’t happen.” DiPietro explained.
These hallucinations cover more than giving customers incorrect facts, this can also include misleading confirmations, inappropriate reassurance, and casual advice that contradicts company policy or regulation.
Furthermore, banking environments can amplify the impact of these unprecedented behaviors since many rely on deterministic logic with high-stakes decisions, strict regulatory requirements, and complex rules.
In these environments, customers and institutions expect a specific, repeatable output for each scenario, as probabilistic systems can create tension by introducing variation where consistency is required.
Scale can also increase the exposure, with even a very small failure rate becoming consequential when interaction volumes rise.
Dan Michaeli, CEO & Co-Founder at Glia, highlights how introducing probabilistic AI into banking creates unacceptable risk, because even occasional uncertainty or error can have serious regulatory and customer-impact consequences.
“It’s high stakes, complex, and regulated, and very often deterministic in its nature,” he said.
“All of the systems that facilitate banking technology have always been very deterministic. To the question, what is my balance? You expect this very specific response.
“You’re applying something that is probabilistic to an industry that is used to being highly deterministic.”
From Self-Service to Skepticism: The Behavioral Fallout of Unreliable AI
Hallucinations can significantly undermine trust in banking because customers expect accuracy when money and long-term decisions are involved.