Velera Finds AI Use Rising as Members Resist AI Transactions

Credit union members appear ready to use AI for information, but not to delegate a purchase or transaction. In Velera’s 2026 Eye on Payments release, 49% of credit union member respondents said they used AI at least a few times per week and three-quarters reported at least one financial use. Yet 59% said they would be uncomfortable allowing an AI tool to make purchases or transactions on their behalf, up from 49% in 2025.

The finding gives payments and digital-banking leaders a useful product boundary. Search, explanation and budgeting support can sit on one side; authority to initiate or approve movement of money sits on the other. A member who accepts an AI-generated answer has not necessarily consented to an AI agent acting on an account.

Velera surveyed 1,750 credit union members and non-members across the United States in July and August 2026. Participants ranged from age 18 to 65 and older, and Velera said the demographic mix aligned with America’s Credit Unions’ Member Profile data. The public release does not disclose the size of the credit-union-member subgroup, response rate, question wording or margin of error. The results are therefore a directional vendor study, not a population estimate or proof that any product design will improve outcomes.

Digital-wallet behavior raises the stakes

The same study found that weekly digital-wallet use among credit union members rose from 11% in 2022 to 42% in 2026. Debit became the most commonly loaded card type, while consumers continued switching among payment methods and wallet cards according to the purchase.

That pattern means an AI feature could influence both the payment rail and the funding source. Product teams should distinguish recommendation, prefilled instruction and completed transaction in their consent design. A member should be able to see what the system proposes, which account or card it would use, any fee or timing consequence, and the exact action that still requires approval.

Revocation matters as much as initial consent. If a credit union eventually permits agent-assisted payments, members need a clear way to narrow or withdraw authority, set limits and review activity. Operations teams also need a reconstructable record showing the member instruction, the system recommendation, the approval event and the final transaction result.

Fraud response remains part of the value proposition

One-quarter of credit union member respondents said they experienced fraud in the past year. Among that group, 52% said the experience increased their trust in their financial institution, and 77% were satisfied or very satisfied with the response. Those are self-reported attitudes, not a measure of loss prevention, but they reinforce the operating importance of communication and resolution after an incident.

An AI payment feature should therefore be evaluated alongside dispute intake, provisional-credit workflows, authentication recovery and contact-center escalation—not only conversion or usage. If a member cannot easily reach a person, identify what the system did or contest an action, the product has moved faster than the surrounding service design.

Use the gap as a test sequence

A practical rollout starts with low-authority assistance: transaction search, plain-language explanation and optional alerts. The next stage can test recommendations without execution. Any move toward transaction initiation should add explicit purpose and limit settings, confirmation at the moment of action, revocation, exception handling and post-transaction review.

That sequence does not assume members will reject agentic payments permanently. It treats today’s evidence honestly: use is rising, while delegated authority remains a separate and more sensitive decision.

Our coverage of agentic-AI shutdown controls outlines the authorization and recovery questions regulators are asking. The member-communications holdout test provides a way to measure whether an AI-assisted message changes behavior without confusing activity with impact.