Five vendor and market signals from the past week, each verified and linkable. Three of the five are fraud and lending plumbing, which continues the pattern we have tracked all quarter: the money keeps going to the back office. If you missed it, Tuesday’s Eltropy vs Glia comparison covers the member conversation side of the vendor map.

1. Navy Federal adds fraud tooling for real-time payments

The largest credit union in the country added AI tooling from Cube to its fraud stack, per American Banker, aimed at scams that ride on faster payment rails. Navy Federal’s fraud operations lead framed it plainly: as real-time payments arrive, the fraud tools have to keep pace. Our take: faster money means faster fraud, and the window for catching a bad transaction shrinks from days to seconds. If your credit union has RTP or FedNow on the roadmap, the fraud tooling upgrade belongs in the same budget line, not a year behind it.

2. PSECU will put a Google Gemini bot in front of members

PSECU, the $10 billion Harrisburg credit union with a half-million members, is working with Candescent to offer a Gemini-based financial assistant inside its mobile and online banking. American Banker’s framing is the interesting part: members already ask ChatGPT and Gemini about their finances, so institutions face pressure to offer their own. Our take: the distribution path matters more than the model. This arrives through Candescent’s digital banking stack, meaning other credit unions on the platform will get the same option soon. Before flipping it on, run it through the member-facing AI evaluation questions; a general-purpose model talking to members about their money deserves the strictest review tier you have.

3. Unit21 puts fraud and AML monitoring natively inside the KeyStone core

Unit21 announced a partnership with Corelation, the core processor behind KeyStone and more than 300 credit union clients. The pitch: detect suspicious activity, act on it in the core, and file on it from a single case. Our take: native core integration is the feature that separates tools that get used from tools that get worked around. For KeyStone shops this collapses the usual integration project. The diligence question is unchanged: ask for two referenceable KeyStone credit unions at your asset size before the demo convinces you.

4. A century-old federal credit union picks Scienaptic for lending

DOCFCU selected Scienaptic’s AI credit decisioning platform to streamline lending and expand credit access. Our take: Scienaptic’s announcement cadence keeps rolling; this is another selection notice, not an outcome report, so apply the usual discount. If lending AI is on your shortlist, our Zest AI vs Scienaptic comparison covers how the two differ and what to demand in writing from either.

5. The cautionary item: the risk is authority, not accuracy

CUToday ran a feature on how risk changes when AI can act, built on a framing from risk intelligence firm FFERM Technologies: the central question is no longer whether the AI gives a wrong answer, but whether it has enough access and authority to act on that answer before anyone catches the error. Our take: this is the cleanest one-sentence risk test we have seen this quarter, and it maps directly onto the control tiers in our model risk piece. Grade every AI deployment by what it can touch and execute, not by how smart the demo looks. An agent that can move money needs a different review than one that drafts an email.

The pattern

Fraud tooling at the largest credit union in the country, fraud and AML inside a major core, lending decisioning at a century-old federal credit union, and one member-facing bot arriving through a digital banking platform rather than a standalone vendor. The back-office spending pattern held again this week, and the one member-facing item came bundled through existing plumbing. The vendor landscape pillar holds our full evaluation kit.

Get this roundup in your inbox every Thursday: subscribe to the AiForCU newsletter.