Board Meeting Digest — July 2025
Artificial Intelligence: Permission, and a Resources Page
Acting Examination and Insurance director Amanda Parkhill and Office of Business Innovation director Amber Gravius briefed the Board on AI. Their message to credit unions was that they may use it. AI is to be treated like any other new product or service: existing laws and risk-management expectations apply, the requirements are largely technology-neutral, and the absence of NCUA guidance on a specific tool does not make it impermissible (the analogy offered: there is no ATM-specific rule either). Staff cited use cases, among them a credit union that cut call volume 40% with machine-learning models and another that screens checks for fraud.
A May 2025 GAO report had found that NCUA's lack of model-risk-management guidance, compared with the banking agencies' 2011 guidance, could disadvantage credit unions. NCUA declined to copy the bank guidance (which one bank trade group had separately called "the most damaging guidance in banking"), concluding that model-risk guidance alone would not cover how credit unions actually use AI; it is instead launching an AI resources page on ncua.gov linking authoritative sources (NIST, CISA). It asked credit unions to flag specific regulatory barriers via the Ask NCUA portal. On its own use, the agency highlighted machine-learning anomaly detection on the 5300 Call Report that saves staff roughly 40 hours each quarter, plus pilots in drafting, translation, cybersecurity, and supervisory stress testing.
In the board discussion, Hauptman called AI one of three generational technologies (alongside the internet and crypto/blockchain) and invoked a "private sector shall lead" ethos; Harper pressed on the promise-and-peril balance, third-party due diligence, and AI hiring (frozen); Otsuka emphasized member impact and data quality. All three made the same accountability point: a credit union cannot blame its AI — "you can't tell the officer your backup camera stopped working." On fraud, the consensus was that AI brings new methods (deepfakes, voice cloning) but not new scams.
The Ombudsman and Duplicative Exam Requests
Ombudsman Shemica Sutton briefed the Board on her office's role and the post-examination survey. In 2024 the office fielded 598 stakeholder inquiries, conducted 10 independent reviews, and made one recommendation to leadership. The survey's 2024 response rate rose to 52% (up 30% from 2023). It also recorded a complaint about process: 41% of participating credit unions disagreed that duplicate requests for data are avoided during exams, with items uploaded to the secure portal requested again and no coordination across examiners.
Parkhill detailed the fix. The chairman's framing was systemic: if enough people do a job wrong, the problem is the job, not the people. Rather than press examiners to coordinate, the agency is changing the mechanics: MERIT-system prompts to grant the whole exam team access, standardized folder structures, a tips page, and potentially AI tooling. ("We have to Netflix this," Hauptman said, meaning find a better way rather than enforce the old one.) Harper raised the still-pending 2012 recommendation to make the ombudsman's office fully independent (stalled by the hiring freeze) and a request for a centralized web page mapping the rules that kick in at each asset threshold ($100M, $250M, $10B). Otsuka flagged a rise in member inquiries about data-breach exposure and share-insurance coverage.
Central Liquidity Facility
CLF president Anthony Capetta gave the second-quarter update. With no major moves in rates, credit conditions, or liquidity availability in 2025, the message was preparedness: contingent-liquidity access matters most before it is needed, and the CLF — credit unions' backstop "discount window," also empowered to lend to the Share Insurance Fund in a system-wide event — is the standing option. Membership remains a short application plus a capital-stock subscription of one-half of one percent of a member's paid-in and unimpaired capital and surplus.QuarterCall Research Assessment
July's meeting carried no data release. We read its AI briefing and the duplicate-request fix as two ways of lowering the operating cost of running a credit union, a cost that weighs most on the smallest institutions. As board member Harper noted, roughly six in ten credit unions hold under $100 million in assets, the same cohort whose earnings the Call Report data show barely clearing breakeven. AI that an under-resourced credit union can rent from a vendor, and an exam process that stops asking for the same file twice, are the kind of marginal relief that helps a sub-scale institution without changing the scale economics underneath.
The accountability point, that a credit union cannot blame its AI, is the supervisory counterweight to the permission: the agency is clearing the path to adopt technology while keeping responsibility on boards and management. We read that as the right posture for an insurer that wants innovation without taking on new risk that no one owns.