Dora NodaOn September 18, 2026, the Office of the Comptroller of the Currency granted conditional national trust charters to three stablecoin firms in a single day — Agora, Bastion, and Catena Trust Bank — the last of which proposes to be a bank for AI agents (Banking Dive, PYMNTS). Sixteen months earlier, Catena Labs had exited stealth with an $18 million seed round led by a16z crypto to build what co-founder Sean Neville — who co-created USDC at Circle — calls the first AI-native financial institution (Cointelegraph). A $30 million Series A in May 2026, co-led by Acrew Capital and a16z crypto, brought total funding to $48 million and an 11-person team (Fortune via Stablecoin Insider). We covered Catena's seed round and Agent Commerce Kit launch when the company emerged from stealth; this post picks up the story with the Series A, the charter, and the compliance market forming around both.
The short version: the GENIUS Act, signed July 18, 2025, turned "issue a dollar token" from a legal gray zone into an engineering checklist — 1:1 reserves, monthly attestations, redemption at par, AML duties, capital rules, and a charter — and the January 2027 effective date put every prospective issuer on the clock. Catena's bet is that the demand side of that market won't be humans at all: autonomous agents that can't hold credit cards at scale will transact in stablecoins, and somebody has to sell businesses a governed way to trust them. This post maps the compliance stack concretely, shows which infrastructure covers each row, and explains what agent builders should take from it.
The GENIUS Act compliance stack, concretely
Strip away the commentary and the law plus its implementing proposals reduce to seven buildable requirements. This is the table every prospective issuer — bank, fintech, or agent platform — is working from:
| # | Requirement | Source | What the issuer must operationally build |
|---|---|---|---|
| 1 | 1:1 reserve backing in high-quality liquid assets (cash, bank deposits, short-term T-bills, Fed balances) | GENIUS Act, signed July 18, 2025 (CoinLaw) | A reserve portfolio restricted to eligible assets, custodied at a qualified institution — no commercial paper, no crypto, no duration games |
| 2 | Monthly public attestations of reserve composition | GENIUS Act disclosure provisions | An attestation pipeline: custodian feeds, auditor sign-off, and published reports on a monthly cadence, not quarterly |
| 3 | Redemption at par, on demand | GENIUS Act redemption obligations | Standing liquidity plus mint/burn plumbing that honors redemptions promptly even in a run |
| 4 | Bank-grade AML and sanctions compliance | BSA obligations applied to issuers, via Treasury-coordinated rulemaking | Transaction monitoring, sanctions screening, and suspicious-activity reporting over on-chain flows — continuous, not retrospective |
| 5 | Capital requirements against credit and operational risk | Fed notice of proposed rulemaking, Sept. 24, 2026 (Reuters) | A capital buffer sized to issuance volume; Agora's conditional charter, for example, requires at least $10M in Tier 1 capital (crypto.news) |
| 6 | A license: federal charter above $10B outstanding, or a federally-certified state regime below it | GENIUS Act issuer tiers (ABA Banking Journal) | Either a charter application of your own (12–18 month conditional-approval clock) or a sponsor/white-label partner that already holds one |
| 7 | Compliance by the effective date | Earlier of 18 months after enactment or 120 days after final rules — January 2027 (CryptoRank) | A program already running, since regulators missed the initial July 2026 rulemaking target and the runway is now measured in months (Crypto Economy) |
Note what this list is: a procurement catalog. Almost nobody launching a branded dollar token in 2027 will build all seven rows in-house. Each row is something to buy, rent, or inherit from a chartered partner — which is exactly why an infrastructure layer is forming underneath the issuers.
The issuer engine, mapped row by row
Here is the same stack with the second column filled in: who actually sells each row today.
| Burden (from above) | Infrastructure that covers it | Who sells it |
|---|---|---|
| Eligible reserves + custody | Reserve custody at a qualified bank; issuance protocols that designate custodians by design | Anchorage Digital (federally chartered) paired with M0 rails (CoinDesk); VersaBank behind Fiserv's platform |
| Attestations + disclosure | Auditor-integrated reporting over custodian feeds | Paxos's trust-company stack; issuer platforms (Bridge, Brale, Coinbase) (DeFi Prime) |
| Mint/burn + redemption | Token lifecycle rails with standing liquidity | Bastion's minting, redemption, and clearing under its conditional charter (AdBytes); M0's configurable issuance protocol |
| AML + sanctions screening | Chain-monitoring and screening wired into issuance | Specialist vendors plus issuer platforms; Catena's planned conversion, clearing, and execution services under its trust charter (Collin Canright) |
| Capital | Balance-sheet capacity | Whoever holds the charter — which is why sponsors, not startups, carry this row |
| The license itself | Charter sponsorship / white-label issuance | Bastion, which issues no stablecoin of its own and exists purely as white-label rails (Tracee); Fiserv's white-label platform, open to ~10,000 bank clients (Odaily); Anchorage's M0-powered issuance platform |
| The deadline | Pre-built, pre-approved paths | The whole category: renting beats the 12–18 month charter clock |
This table is the answer to the scale question. The stablecoin market stands at roughly $300 billion — $300.9 billion across 195 assets, 152 issuers, and 47 chains as of October 1, 2026 (RWA Foundation via CryptoCompass) — yet two issuers hold about 83% of it (CoinLaw), with Tether alone near 60%. The a16z thesis behind Catena is that this concentration is a supply artifact, not a law of nature: issuing used to require becoming a money-transmitter-licensed, audit-retained, bank-partnered institution, so only a handful of firms could do it. White-label rails plus a replicable charter path collapse that fixed cost, and the market goes from a handful of issuers to hundreds — banks, telcos, marketplaces, agent platforms — each with a branded dollar token riding shared compliance plumbing.
Concretely, a fintech launching in this window follows three steps: first, charter-or-sponsor (apply, or plug into Bastion/Fiserv/Anchorage/Paxos and inherit theirs); second, rails integration (mint/burn, custody, wallets — M0-style protocols or a white-label API); third, reserves-and-attestation wiring (eligible assets at a qualified custodian, monthly auditor loop). Firms that complete step one own the relationship; firms that stop at step two rent it. Catena's charter application is a bid to own it.
Catena's bet, piece by piece
The funding timeline tells the story in three acts. Act one: the $18 million seed on May 20, 2025, led by a16z crypto with Breyer Capital, Circle Ventures, Coinbase Ventures, CoinFund, Pillar VC, Stanford Engineering VF, and Tom Brady participating — a nine-person team building an "AI-native financial institution" for agents that hold assets and transact at scale (Gate Learn, The AI Insider). Act two: the $30 million Series A in May 2026, co-led by Acrew and a16z crypto with Oak HC/FT joining, alongside a Private Access program — total raised, $48 million (Forkast). Act three: the September 18, 2026 conditional approval of Catena Trust Bank, covering custody of payment stablecoins plus conversion, clearing, and execution services — one of roughly a dozen conditional trust charters the OCC has granted digital-asset firms since late 2025, following Circle, Ripple, BitGo, Fidelity, Paxos, and Coinbase (VitalLaw, RefreshCoin).
The product thesis is Neville's sharpest line, given to Fortune at the Series A: "giving an agent a wallet is pretty easy compared with giving a business a governed way to trust it." Catena's governance layer for agent transactions is exactly that second half — spending limits, approved recipients, deterministic policies, and audit trails sitting between an autonomous agent and real money. It is the financial equivalent of an admission controller: the agent proposes, policy disposes, and every decision lands in an audit log.
Notably, part of the stack is open source. Catena's Agent Commerce Kit (ACK), released under MIT terms at launch, defines protocols for agent identity (ACK-ID, built on W3C decentralized identifiers and verifiable credentials) and payments (ACK-Pay, a transport-agnostic payment pattern), so agents can operate their own accounts and wallets against standard rails (Yespress). Open protocols for the edges, a regulated bank for the money: that split is deliberate, and agent builders should read it as an invitation to build on the open half today.
The competitive map: who sells "issue your own dollar"
Catena is not competing with every row of the issuer-engine table — it is betting the demand side (agents) while others fight over the supply side (issuers). The supply-side field, ranked by how directly each firm sells issuance:
- M0 ("M-Zero") — a protocol for minting fully configurable stablecoins with custom branding, compliance rules, and reward logic; its partnership with Anchorage Digital in April 2026 married the protocol to a federally chartered custodian, and integrations with Stripe, MoonPay, and MetaMask put it inside existing distribution (CoinDesk).
- Paxos — the trust-company incumbent: issues others' dollars (including PayPal's) under New York oversight and now holds a conditional federal charter, the model every white-label rival is measured against (CoinLaw).
- Agora — issuer of AUSD since August 2024; its conditional charter lets it move issuance from a Bermuda affiliate into a U.S. national trust bank, the onshoring play in miniature (BigGo Finance).
- Bastion — pure infrastructure: custody, wallets, clearing, and white-label issuance, with no proprietary stablecoin competing against its own customers (CryptoCoverage).
- Fiserv — distribution at a scale no startup can match: a white-label digital-asset platform offered to roughly 10,000 financial-institution clients, with minting, custody, and reserve management handled by OCC-chartered VersaBank (Odaily).
- Mento and Angle — the decentralized flank: Mento's FX protocol processed $18.5 billion in 2025 volume across local-currency stables and expanded to Polygon in July 2026 (Cointelegraph), while Angle's dollar-backed USDA pairs RWA backing with on-chain forex ambitions (Crypto Economy). Neither targets U.S. chartered issuance — which is precisely why GENIUS leaves them a lane.
Catena's differentiation is directional, not incremental: everyone above sells to issuers, while Catena is building for the transactors — and specifically for transactors that are software. If agents become meaningful economic actors, the issuer-side bloodbath over fees and float is someone else's problem; the scarce asset is the governed account the agent spends from.
What it means for builders
If you are launching a token, distill sections two and three into a launch checklist: confirm your charter path first (own application vs. sponsor — the 12–18 month clock decides your 2027 feasibility); pick rails that already designate qualified custodians (renting M0/Anchorage/Bastion-shaped plumbing beats negotiating custody yourself); lock the monthly attestation loop before launch, not after; and budget Tier 1-style capital from day one rather than discovering it in diligence. The upshot is that none of this is novel engineering anymore — it is vendor selection plus a lawyer.
If you are building AI agents, the takeaway is architectural. Agents cannot use credit cards at scale — cards assume human cardholders, chargeback processes, and fraud departments, none of which survive contact with autonomous software spending programmatically (Yellow). Stablecoins fit because they are programmable, settle fast, and cost fractions of a cent — adjusted on-chain stablecoin volume ran to $10.2 trillion over the trailing twelve months on Visa's dashboard (CoinLaw). What has been missing are the two primitives Catena is productizing: portable agent identity (DIDs and verifiable credentials, via the open-source ACK you can adopt without permission) and deterministic spending governance (limits, allowlists, audit trails) that lets a business hand an agent a budget instead of a blank check. Build your agent's wallet against open protocols, and keep the policy layer — who may spend what, where, with what approval — as explicit, reviewable configuration, the same way you would treat deploy permissions on infrastructure you own.
Open questions and risks
Three caveats keep this story honest. First, conditional is not open: Catena's approval expires unless capital is raised within 12 months and the bank opens within 18, and the OCC has been explicit that business-plan departures need prior notice — the charter is a funded milestone, not a done deal (Stablecoin Insider).
Second, the rules themselves are still proposals: the Fed's September 2026 capital and reserve-diversification details, the Treasury-coordinated AML rulemaking, and the state-regime certification process all have to finalize before January 2027, and regulators already missed the first deadline. Third, concentration cuts both ways: a market where two issuers hold 83% may simply not fragment, if network effects in liquidity and acceptance outweigh the cost savings of white-label issuance. The few-to-many thesis is a bet on plumbing beating network effects — plausible, unproven.
Compliance becomes software; customers become agents
Step back and the shape of the trade is clear. The GENIUS Act converted regulatory permission into a bill of materials, and a whole vendor stack — protocols, custodians, white-label banks, attestation pipelines — now sells that bill of materials by the row.
Catena's $48 million wager adds the second half of the convergence: while compliance becomes software you procure, the customer on the other side becomes software too. Somebody has to bank the agents. Whoever sells businesses the governed account — identity, limits, policy, audit trail — owns the demand side of the dollar-token economy the way the issuers own the supply side.
Bex.co is built for the same governed-autonomy future on the infrastructure side: AI agents that deploy and operate apps need explicit permissions and reviewable policy, not blank checks — on machines you own. Star the repo on GitHub or deploy your first app today.
