October 6, 2026
Four stablecoin developments bankers should know about

The Stablecoin Banker is a periodic newsletter keeping bankers on top of the stablecoin industry. I highlight top stories that are relevant to banks, with my insights and commentary to draw out the most important conclusions.

If you find this content useful, please feel free to forward it to your friends and colleagues. They can also subscribe directly or read back issues at thestablecoinbanker.com. If you're interested in stablecoin services for your bank, feel free to reach out to us at Omnia.

"If every household used AI agents to optimize the return on their cash balances, banks could lose a large share of the cheap deposits they rely on to make loans."
— Torsten Slok, Apollo Chief Economist, "Is an Agentic Bank Run Coming?" (Sept. 27, 2026)

In This Issue

• The Fed Floats a Stablecoin That's Cheaper to Hold Than a Deposit
• The De Novo Pile Is a Breakup Letter, Not a Crypto Story
• Coinbase Will Happily Be Your Back Office. It's Also Reading the Mail.
• The Agentic Bank Run Is Oversold. The Agentic Checkout Isn't.

Plus, tidbits you may have missed in our Coupon Clippings section.

The Fed Floats a Stablecoin That's Cheaper to Hold Than a Deposit

On September 24, the Federal Reserve Board requested comment on two proposed rules implementing the GENIUS Act for the payment stablecoin issuers it supervises (state member banks, and uninsured state-chartered issuers with $10 billion+ in issuance). The proposals set reserve, capital, risk management and custody requirements, amend the Board's capital rule, Regulation Q, and create an application process for state member banks. Comments are due November 30, 2026. Under the proposal, issuer required capital would be 2.0% of its first $20 billion outstanding, 1.5% of the next $30 billion and 1.0% above that. Question 231 asks about the appropriate risk weight a bank should apply to its holdings of payment stablecoins. The suggested level ranges from 100% for issuers with non-issuance activity greater than 5% of its assets or liabilities, 20% for those with greater than 5% of stablecoin reserves in uninsured deposits, and "five to ten percent" for pure-play issuers with low uninsured deposits (insured deposits and Federal Reserve balances do not count against an issuer). The OCC, FDIC and NCUA proposals issued earlier this year do not address how a bank should risk-weight a stablecoin it holds. Sources [Federal Reserve, Federal Register]

My Take:

The Fed is the first banking regulator to put numbers on the risk a bank takes holding a GENIUS stablecoin, and its numbers are lower than mine. The passage is a question, not rule text, and risk weights normally arrive as a three-agency rule. But Regulation Q reaches every bank holding company regardless of the actual bank's chartering status, so the Fed's answer has an outsized impact.

In Banking Stablecoins, we argued a coin from a national trust bank is an exposure to a depository institution and takes the same 20% weight as a deposit at another US bank, while a corporate issuer's coin is a 100% corporate exposure. In March I wrote that regulators would find look-through treatment at 0–20% the easier argument. Footnote 153 draws the same fork, and Question 231 goes further: 5% or 10% for a near-pure-play issuer holding 5% or less of its reserves in uninsured deposits, and 100% for an issuer that runs other business. The low tier beats a deposit at another bank, which hands bankers an awkward fact: the coin that treats a bank's balance sheet best is the coin whose issuer keeps its reserves out of banks.

The NPRM never says look-through, but the logic is one. Treasuries carry 0%, so the Fed prices only the "minimal but still existing residual credit risk" of the issuer, which grows with uninsured deposits (Circle's $3.3 billion at SVB is cited by name). Today's large issuers hold meaningful bank deposits, and since they don't disclose whether their deposits are kept in a FDIC sweep program, we don't know the deposit split between insured and uninsured. That said, many issuers would probably land at a 20% risk weight.

Whether issuers can actually meet the 5% threshold requirements implicates intraday liquidity. By my rough math, insured sweep networks at about $200 million per depositor plus a Fed payment account give an issuer on the order of a billion dollars of liquidity that stays off the 5% line. The OCC's proposed safe harbor asks for 10% of reserves in demand deposits, insured deposits or Fed balances; for a $50 billion issuer that is $5 billion, most of it necessarily uninsured. The two proposals conflict only above a certain size, and that size is where the volume sits.

Otherwise the Fed NPRM tracks the OCC and FDIC, apart from a formulaic capital charge of up to 2% of issuance in place of a principles-based test. A 15% return on equity then needs about 30 basis points of retained yield, ten times the OCC's proposed variable capital charge. Feasible, but it forces the issuer to constantly raise equity to grow.

For bankers: with additional clarity on the risk weight, holding a stablecoin on your balance sheet becomes even more feasible. In fact, even at the 20% level, the cost to hold the stablecoin likely comes in below the cost to prefund a balance at a stablecoin provider (which would receive the 100% risk weight treatment as a corporate exposure), the preferred route for banks so far to keep stablecoins at arms length.

The De Novo Pile Is a Breakup Letter, Not a Crypto Story

On September 18 the OCC granted conditional approval to three uninsured national trust banks: Bastion (a conversion of its New York trust charter; stablecoin custody, wallets and white-label issuance, including Sony Bank's coin), Agora (which will move issuance of its AUSD stablecoin from Bermuda into the bank) and Catena Labs (custody, conversion and clearing). Block said on September 9 that it had filed for Builders Bank & Trust, N.A., an uninsured national trust that would custody bitcoin and stablecoins for Cash App and Square customers, and replace more than 50 state licenses. Avant filed on September 18 with the OCC and FDIC for an insured Avant Bank, N.A., moving its loan and card programs off WebBank and adding checking and savings; CEO Al Goldstein said the charter would "help reduce our cost of funds, and enable us to offer products under one regulatory framework." As of August 5, the OCC counted 40 de novo applications since January 2025 and 21 approvals, and said it has decided many "within 120 days." Sources [American Banker, Banking Dive, OCC]

My Take:

The de novo pile often gets read incorrectly as a crypto story. Rather, it is a catalog of ways to run financial services without renting a depository bank, and sorted by business model the crypto-only applicants are a minority.

The charter map, by business model (status as of early October):

Start with the bucket everyone has in mind: Coinbase, Kraken and Crypto.com, plus Fidelity and Morgan Stanley. None of these applications expands what the applicant can do under state money-transmitter licenses. The charter buys an exit from the 50-state MTL regime and, potentially, from state crypto rules like New York's BitLicense, with its 12-month-plus waits. The OCC has put that in writing: Interpretive Letter 1192 holds that state licensing is preempted for a national trust bank even where a state limits its exemptions to a subset of national banks. However, the BitLicense rule exempts only New York-chartered institutions, and NYDFS has said nothing which may tee up a later conflict.

Stablecoin issuers get the same preemption, plus the expectation that national banks will have the easiest path to a GENIUS permitted-issuer license next year. However, issuing through a trust charter today has similar operational costs to doing so under an MTL because the trust still relies on banking partners for deposits and payments, carries the same or stricter compliance obligations, and may need to carry more capital. Direct Federal Reserve access changes that. Kraken Financial, a Wyoming-chartered fully-reserved special purpose bank, got a master account from the Kansas City Fed in March in a format that, as I hear, resembles the payment account the Fed proposed in May. An issuer with that account can decouple, in whole or in part, from the depository system.

The payments applicants are the most interesting, because they are a step toward unbundling payments from banks. To make a payment in the United States, funds must sit on deposit with a bank, directly or through a fintech. These applicants can already offer stablecoin balances, pay rewards on them (pending final issuer rules on yield), and enable spending via cards settled directly to the networks in stablecoin: a basic depository product. In this bucket, bank payments flow through correspondents for now, until the Fed payment account comes into existence.

The BaaS exits tell us that at sufficient scale, fintechs want their own charter. Sponsor banks are slow to onboard, run antiquated APIs, cost too much, and get cold feet. Lincoln Savings Bank, the sponsor behind Acorns, Qapital and others (and historically Cash App), is the current example: its merger agreement with Equity Bancshares expects a "Complete Exit" from its sponsor platform, bars new contracts, and raises the purchase price by $750,000 if the exit is done by December 31. That gives its fintech programs merely four months to move their business - a kind of risk that justifies a captive charter. The applicants should still study SoFi, where, as I understand it from former employees, becoming a bank dramatically changed culture and velocity; segregating the bank from the rest of the business will matter.

I've skipped the full-service bucket for brevity, but most of the above applies there too. The common thread is decoupling from traditional banking, by reshaping what a depository bank is or by pulling whole business lines out of one. Incumbent banks can cut prices to compete with new models which may be more cost effective, but cost is not the main motive. Applicants want control of decision making, a modern product and a business built for their segment. Delivering those changes requires rethinking culture, technology and go-to-market in ways that are foreign to most banks.

Banks should read the payments bucket as the new competitor set for the bank's payments business. A balance, rewards and a card with no deposit account underneath is a product a trust company can now ship, and matching the capabilities that fintechs have demanded - and can now access on their own - is a decision for the top of the house, not the pricing committee.

Coinbase Will Happily Be Your Back Office. It's Also Reading the Mail.

Coinbase announced two bank distribution deals in September. Coinbase and Stablecore said they will bring buying, selling, holding, staking, custody and stablecoin payments to the 3,000-plus US banks and credit unions reachable through Stablecore's integrations with Jack Henry and Q2. Here, Coinbase supplies the custody and exchange infrastructure, the services are white-labeled inside the bank's own digital banking, and the bank keeps the customer relationship, with Amarillo National Bank in Texas the first institution named as having started work. Citi and Coinbase expanded a collaboration that began in October 2025. Coinbase will use Citi's Virtual Account Wallet, part of Citi's Banking-as-a-Service suite, to power Coinbase Virtual Accounts. Those accounts offer fintechs and businesses the ability to hold stablecoin balances with immediate on- and off-ramp through Citi rails. Separately, Citi institutional clients using Spring by Citi, the bank's merchant acceptance platform, will be able to accept stablecoin at checkout, with Coinbase Payments converting the stablecoin to fiat and Citi settling the funds as bank of record. Sources [Citi, Coinbase]

My Take:

Brian Armstrong said a year ago, "Ultimately, we want to be a bank replacement for people. We want to be their primary financial account." Coinbase nonetheless keeps signing distribution deals with the banks it means to replace, and it walks the friend-or-foe line by selling each bank only the parts of its business that stay clear of that bank's deposits.

Citi is one version of the compromise. Virtual Accounts are Coinbase's product, sold to the fintechs on its platform; Citi is the plumbing and is not promoting them to its own customers. Spring converts stablecoin to fiat before the merchant sees it, so the money lands as a Citi deposit, and no Citi depositor gets a path into stablecoin. The dilemma sits one level up: a corporate holding working capital in a Coinbase Virtual Account as stablecoin has replaced an operating account at some bank. That product is built to compete with bank deposits generally; Citi just gets paid when the product moves money over fiat rails.

Stablecore demonstrates a different compromise, white-labeling Coinbase custody and exchange inside the bank's own online banking so the bank keeps the relationship. That helps the competitive dynamic more than it helps adoption: the model typically means prefunding at Coinbase or customers waiting for funds to land there, the tradeoff I covered two weeks ago on the Moov deal and in Banking Stablecoins. Simpler integration doesn't shrink the work of launching a bank product, and demand still looks thin to the average community banker.

Banks shouldn't avoid Coinbase for now; its infrastructure is tested over a decade. The catch is quieter. Coinbase's developer docs have white-label partners onboarding users to Coinbase behind the scenes, with Coinbase running KYC, so even under a bank's label Coinbase gets to see every trade and payment on its systems. I suspect that view, pooled across thousands of banks, shows Coinbase exactly which segments want a primary financial account elsewhere, and nothing in the public terms of service stops it from marketing to them directly.

For bankers: before signing a white-label digital-asset deal, write data-use and cross-marketing restrictions into the contract: what the provider may do with the identity, trade and payment data it sees, and a bar on marketing to the bank's customers with what it learns from the bank's traffic.

The Agentic Bank Run Is Oversold. The Agentic Checkout Isn't.

On September 27, Apollo chief economist Torsten Slok published a note titled "Is an Agentic Bank Run Coming?", arguing that assistants like Meta's Muse "could soon sweep household cash automatically into accounts paying 3.3% to 5.0%, instead of the 0.1% national average on checking accounts," and that banks "could lose a large share of the cheap deposits they rely on to make loans." The next day Instinct, an SMS-based personal agent that books travel, makes reservations and buys goods for its users, raised a $1 billion Series C at a $10 billion valuation led by Sequoia, Benchmark and Coatue; founder Noah Shinn said that 40% of users have shared a personal credit card with the agent by their third week of membership. Both Instinct and Muse, which Meta launched September 8, pay through Stripe's Link wallet, which issues a single-use virtual card for each purchase the user approves; Amazon began blocking Muse from its site on September 20. On September 22, Bank of America, Capital One, ING, NatWest, Commonwealth Bank of Australia and ASB published voluntary principles for "trusted agentic commerce" covering transparency, safety, privacy, choice and interoperability. Sources [Apollo, TechCrunch, American Banker]

My Take:

The latest wave of AI hyper-freakout arrived with Apollo's claim that agents will optimize everyone's bank account. The assertion is simplistic, and critics quoted by American Banker, from Morgan Stanley's former head of AI to the consultants at Simon-Kucher, dispute the speed at which any of it could happen. I wouldn't write it off. Whenever I think "AI can't do that," I remind myself where AI was twelve months ago.

In the past, I have been skeptical of agents paying with cards. The value of agentic shopping is in the search, the all-in price comparison and the choice; the purchase itself seemed trivial, since the agent can load the cart page, hand it to me, and my password manager can fill in the card in my browser. My belief was changed when, few days ago, I shared my card with Instinct. My logic was not wrong, but it was incomplete. In the human world, shopping may be 95% of the effort of the buy flow. Once the agent reduces shopping to a few text messages, paying becomes the largest remaining chore, and my strong preference is to text back "go buy it." The implementation is also less risky than I expected: my card sits with Stripe, not the chatbot, and Stripe pushes a notification and mints a one-time card for each purchase. Banks, networks and large fintechs are now working out liability rules that could automate even that step.

The open question is whether agents will sweep household cash into high-yield accounts, as Apollo suggests. Agents are good at exactly this kind of work: patiently slogging through websites and forms. The hurdles are trust and adoption. A user has to hand over far more than a card number, and Oppenheimer finds only 8% of consumers would trust Meta with their passwords, while Google garners only 30%. Banks may react as Amazon did and block the agents, though that path is contested; Amazon's injunction against Perplexity's shopping agent was thrown out on appeal in August. The research on depositor inattention (Lu and Wu's "Banking on Inattention," out this year) says people don't switch banks because they aren't thinking about it, rather than because of the friction of switching. For agents to change that, consumers first have to adopt them, then trust them with the information to execute a switch the agent proposes at a non-creepy moment. I think it happens, after much broader agent adoption.

Tokenization sits behind both halves. Tokenized deposits make moving money between accounts fast enough to reward switching and drain low-yield balances, and stablecoins extend agentic purchasing to small-dollar, just-in-time digital services.

All of this is early (Instinct recently reported only 100k users), but banks are less accustomed to exponential change than most sectors. The problem is that you often can't see exponential progress until you're looking over your shoulder. Banks should consider building the cheap option now, as I argued in June: stablecoin and wallet capability your customers' agents can draw on. On the deposit side, watch adoption metrics like consumer adoption of interactive AI agents (not just chatbots) and the volume of payments running through Stripe Link and similar products. Rising adoption should trigger a proactive response before it goes exponential.

Coupon Clippings

OUSD Goes Live, With Five Hands on the Wheel

OUSD, the Open Standard consortium's dollar stablecoin, went live with Stripe-owned Bridge as the issuer, reserves held at BlackRock, BNY and Lead Bank, and more than $1 billion in launch liquidity committed by the five founders. Open Standard will be governed by a board "eventually drawn from its founders" (Coinbase, Mastercard, Shopify, Stripe and Visa), while distributors earn reserve income "in proportion to the supply and activity they drive" and can earn equity in the company. The management fee, the yield formula, the equity threshold and the circulating supply remain undisclosed. In July, I wrote that shared-ownership issuance is the only way to challenge Circle and Tether, and that governance would decide whether OpenUSD could pull it off. Control now sits with five institutions rather than the 200-plus on the original press release, which is a risk, but those five have a lot to gain and little to lose from stablecoin adoption, so the interests are aligned. Offering equity to additional members for demonstrated distribution is the right structure to encourage productive participation, even if the economics a distributor would need to evaluate aren't yet public. Demonstrating a different model of paying for distribution, Circle also sold to Binance $100 million of stock at a 5% discount and renewed a five-year deal paying it incentives tied to USDC balances. Note how equity in the issuer shows up in both models. Sources [Bloomberg, Circle]

The SEC Builds a Stock Market Without an Exchange

On September 17 the SEC issued an innovation exemption that lets "Tokenized Securities Venues" trade tokenized NMS stocks without registering as an exchange or ATS, using automated market makers on public blockchains. The exemption limits implementation to allow-listed participants; trading pairs with tokenized stock, a payment stablecoin or a tokenized money market fund; a 0.25% volume cap; and, zero margin trades. Third parties may tokenize a company's shares, but the issuer gets 30 days' notice and a veto, and the order expires in 2031. The exemption opens the door to a new model of equity market access that should interest banks: trading moves off infrastructure owned and run by the exchanges and onto venues anyone can operate, with settlement in participants' own wallets and stablecoins in the cash leg by design. A bank's path in doesn't run through a trading desk. Its customer converts deposits into stablecoin and carries that stablecoin into the venue, so the bank that already offers the conversion becomes the on-ramp to the stock market without building a brokerage. For now the venues are a sandbox, capped at a rounding error of daily volume and not yet open for business, but the architecture is the point. [Full Story]

Thanks for reading. If you don't hear from us for a bit, don't worry—we'll be back when there's something important to share.

Omnia is a provider of stablecoin infrastructure for banks that want to capture growing demand for stablecoins. If you're interested in learning more about us, please get in touch.

Getting value? Forward to a colleague and subscribe.