Good morning!
Carey Ransom said something on the podcast a while back: "Assume AI can do everything, then prove to yourself why it shouldn't." Feels a lot less hypothetical this week. FDIC just approved a bank built to run at AI speed instead of human speed.
Let’s get into it.
TRENDING AI NEWS FOR CB
FDIC approved a bank built for AI agents, not humans
Augustus National Bank got the FDIC's sign-off this week to launch as a "programmable dollar" bank, built for machine-speed settlement instead of the two-day clearing cycle most correspondent banking still runs on. It'll serve financial institutions, high-net-worth individuals, and tech companies with stablecoin issuance, custody, and payment services.
Why it matters for your bank: the assumption that a human is checking out is already broken. Banks that wait for their fintech partners to force the agentic-commerce conversation will be reacting instead of leading it.
AI is now a standing question in every bank exam
OCC and Federal Reserve examiners have begun inquiring about AI governance, vendor risk, and emergency shutdown capabilities in every routine exam; no bank review is exempt from this requirement anymore. The three agencies also revised their model risk management guidance in April, though it explicitly doesn't yet cover generative or agentic AI.
Why it matters for your bank: there's no dedicated AI rulebook yet, so examiners are stretching the frameworks you already use. "We haven't gotten to that" is not going to hold up much longer.
The Community Bank Leverage Ratio just dropped from 9% to 8%
FDIC, OCC, and the Fed finalized the lower CBLR threshold this summer, and the compliance guide update lands now, with the grace period extended from two quarters to four.
Why it matters for your bank: more banks under $10B now qualify for the simplified framework, freeing up capital planning hours that could go toward AI governance instead of leverage-ratio paperwork.
DEEP DIVE
A Bank Built For Robots Just Got Approved. Who's Liable When It Goes Wrong?
Augustus just got approved by the FDIC. It's a real bank now, and it's built to move money between AI agents, not people.
Think about what your fraud model actually assumes right now. A human customer on one side. A human employee catching the weird cases on the other end. That's baked into your AML controls, your fair lending checks, all of it. Augustus just proved that assumption doesn't hold anymore.
So what happens when one AI agent deals with another AI agent, and it goes wrong? Nobody has a clean answer yet. Experian's calling this machine-to-machine fraud, where it gets murky fast, who's actually responsible once an agent does something nobody signed off on.
You don't have to be banking AI agents for this to matter to you. If you're already using AI for document review, fraud triage, loan servicing, whatever, you're building toward the exact same problem. The smart move isn't waiting around for a lawsuit to sort out who's liable. Write down now, on paper, what a human actually has to check before an AI action is final.
Three things to do this week:
Ask whoever owns your AI vendor relationships whether any tool you use can act autonomously across systems, not just answer questions. That's the line that matters.
Write down, on one page, where human sign-off is mandatory before an AI-touched decision becomes final. If that page doesn't exist, you don't have a policy; you have a hope.
Ask your fraud and compliance teams whether your current controls assume a human on both ends of every transaction. If yes, start planning for the ones that won't.
FROM MULTIMODAL
Welcoming Aaron Cain to the Multimodal Advisory Board
Aaron Cain Joins Multimodal Advisory Board — Credit Union 1 Digital Innovation
Aaron Cain leads digital innovation at Credit Union 1, and his focus sits squarely on the customer-facing side: the interfaces, the channels, the everyday digital experience people actually touch.
We wanted that lens on the board on purpose. Business innovation and digital innovation don't always agree on what matters first, whether you're a bank or a credit union. We'd rather have that tension sitting at the table than smoothed over before anyone sees it.
Welcome, Aaron.
If this week's issue has you thinking about where your bank actually stands on AI governance, book a few minutes for a walkthrough
The Quick Check
Do OCC and the Fed require a bank AI policy?
Not as a standalone rule, not yet. But both agencies have made AI a standing topic in every routine exam. No review skips it now, rule or no rule.
What do bank examiners actually ask about AI?
Three things, consistently: how is the AI governed, what technical limits keep its behavior in bounds, and does a human review its output before it becomes a final decision. Increasingly, they're also asking whether an emergency shutdown exists.
Does a small community bank get the same scrutiny as a big one?
The guidance says practices should scale to size and complexity, so the depth of the answer can flex. But no bank, regardless of size, gets to skip the question entirely anymore.
🎁 SEND THIS, GET THIS
Know someone still doing this by hand?
Forward this issue to one person on your team who's still buried in manual document review, and we'll send you "The True Cost of Manual Document Processing in Credit Unions." Yes, it's credit union data. But underwriting, compliance, and account-opening paperwork cost the same to process by hand whether you're chartered by the NCUA or the OCC. The benchmarks translate directly.
Takes ten seconds. Beats another Slack message about it.
That's this week
See you next Thursday.
