For years the stablecoin debate has been argued from the rich-country side. Will dollar tokens replace card networks, compete with bank deposits, move Treasury markets. This month the IMF looked at it from the other end, from the countries that receive these dollars rather than issue them, and the picture is very different. The paper is Foreign Currency Stablecoins: Initial Considerations on Policies to Manage and Enhance Macrofinancial Stability. Careful, two-sided, and it carries the most useful cost measurement published about stablecoin payments this year. Start with a framing that sounds obvious once said. Stablecoin market capitalisation went from about $6 billion at the start of 2020 to $317 billion in June 2026. USDT and USDC are roughly 90% of it. From the point of view of nearly every country on earth, 99% of stablecoins outstanding are issued abroad in a foreign currency. The euro stablecoin market was about $680 million in June. So for most central banks the question is not whether to issue one. It is what happens when residents hold someone else's. Then the flows. The Fund estimates about 90% of stablecoin activity was cross-border in 2025, and gross cross-border flows of USDT and USDC reached $6.3 trillion that year. Placed next to a global cross-border payments market of around $1 quadrillion, that is small, and the paper says so. What is not small is the shape. On SWIFT, customer payments between advanced economies are the largest share by value and flows between emerging economies the smallest. For stablecoins in 2024 it was almost inverted: 34% between emerging and developing economies, 52% between those and advanced economies, only 14% between advanced economies. 86% of the flows the Fund measured had an emerging or developing economy on at least one side. The rail the rich-world debate treats as a challenger to card networks is, in practice, mostly being used somewhere else. Relative to GDP, the largest holdings in 2025 were in Venezuela, Ukraine and the Kyrgyz Republic. The Fund puts the caveat in a box of its own: activity spans more than thirty blockchains, addresses carry no residency, and every allocation method rests on assumptions. No single method is fully accurate. The caveat travels with the number. Now the part that should change how this industry sells itself. The annex reviews a study by Reslow and co-authors covering 270 directed corridors, 18 countries, 29 exchanges and more than 4,000 complete routes. Assuming the cheapest available ramps, a $200 transfer costs about 2%, below the G20 target of 3% and below the global average for traditional remittances. Then the qualifications. In roughly one third of matched corridors, the cheapest traditional provider is still cheaper. Trading fees and fiat deposit and withdrawal charges are most of the cost. Blockchain transfer fees are less than a quarter of it. And the stablecoin premium, the gap between the local price of the stablecoin and the benchmark rate, pushes the cost below zero in some corridors and above 8% in others. The chain is no longer the expensive part. Getting into the dollar and getting out of it is. Anyone selling stablecoin payments on a cheap transfer fee is selling the quarter of the cost that was already solved. The paper is two-sided and should be read that way. Real benefits: cheaper and faster cross-border payments, competition, programmability, inclusion. Real risks for the countries on the receiving end: currency substitution, disintermediation of domestic banks, weaker monetary transmission, reduced effectiveness of capital flow measures, and loss of seigniorage to foreign issuers. Its first line of defence is sound frameworks, comprehensive regulation, better data and better domestic payment systems, not blanket bans. A payment network that carries instability into the countries it reaches is not one anyone should want to run. Which tells you where the value sits. The Fund notes stablecoins used for payments are under 1% of stablecoins outstanding. The open problem is acceptance, a business that receives the dollar, settles in it and pays onward without converting out and back in. Every payment received in $USDC and paid onward in USDC removes an off-ramp and an on-ramp from the cost stack. On @arc , @circle Layer 1, gas is paid in USDC and finality lands under a second, so the asset you are paid in is the asset the network charges in. @Meridian_Fi Pay accepts USDC and EURC at checkout, gasless for the payer. YieldVault wraps Circle USYC so a waiting balance earns Treasury yield, from the underlying Treasuries, no fixed or guaranteed rate. PaymentStream pays by the second. @FirmataProtocol carries identity, reputation and conditional settlement for agents. Circle has also announced StableFX for on-chain conversion between stablecoins like USDC and EURC. Arc testnet, chain 5042002. +20 contracts verified on Arcscan, Not mainnet. An independent audit of the contracts and the front end gates that migration and it has not happened yet. The cost to beat is the ramp, not the chain. Full piece: community.arc.io/home/forum/… Not investment advice #Stablecoins #USDC #Payments

The most successful stablecoin product of the last two years is one the merchant never sees. A cardholder taps five dollars at a coffee shop. The balance behind the card sits in USDC. In the roughly 400 milliseconds before the terminal confirms, an issuer processor checks the account has the funds, and then the payment settles exactly like every card payment. The coffee shop receives local currency, a few days later, through the network it has always used. Nothing on its statement says stablecoin. F-Prime, the venture investor, mapped this market at the end of August, and by any measure it is working. @Visa estimates stablecoin-linked card volume grew 319% to $5.2 billion in 2025, across more than 130 card programs in over 50 countries, as cited by F-Prime. Paymentscan puts cumulative spending on stablecoin-powered cards above $10.9 billion, and RedotPay forecasts annual spending could reach $50 billion by 2028, both as cited by F-Prime. The reason fits in a sentence. For somebody holding digital dollars, a card is the shortest path from a balance to groceries and fuel. And unlike the first crypto cards, the holder is not liquidating a volatile asset at the till. F-Prime notes Coinbase has charged 2.49% on purchases funded with non-stablecoin crypto. A dollar stablecoin behind the balance removes most of that friction. Then the line that reframes the whole category. @raincards 's chief executive, as cited by F-Prime, said stablecoin-funded payments have reached more than 100,000 merchants without them knowing it, because the transactions still settle through traditional payment networks on T+3. A hundred thousand businesses have accepted stablecoin-funded payments. Not one of them chose to. Not one received the settlement speed, the readable record or the currency choice the payer's money carried right up to the moment it touched the card rail. That is not a criticism of the card model. It is the reason the model scaled. Card acceptance already exists nearly everywhere a business takes payment, and a product that rides it inherits that reach on day one. But it means the industry built a remarkable consumer product on top of a merchant experience that has not moved. F-Prime says it directly: the next step is moving stablecoins deeper into merchant settlement. The merchant side is the harder half, for a structural reason. A consumer product needs one decision, the holder's. A merchant settlement product needs the finance team, the accounting system, the treasury policy and often the bank to accept a different form of money on the receiving side. F-Prime's own map shows where the stack is still thin: round-the-clock settlement platforms, crypto-native regulated banks, and stablecoin-native issuer processors, which it says no provider has yet established at scale. So the useful signal this month is not another card launch. It is where settlement is starting to appear. On 7 October, @circle and Tereina, a financial services company backed by SAP, announced an integration bringing USDC and EURC into enterprise payment workflows, beginning with SAP Pay, with @arc as the preferred blockchain. Eligible businesses would send and receive stablecoin payments without leaving the applications they already use. The announcement describes the SAP ecosystem as generating 84% of global commerce, which is SAP's figure. Proof-of-value pilots come first, over the coming months. That design choice is the important part. It does not ask a finance team to adopt a new system. It puts the settlement option inside the system the finance team already runs. The consumer side reached merchants by hiding inside the card rail. The settlement side will reach merchants by showing up inside the software that already holds their invoices and payables. Only the second delivers the money in the form the payer sent it. What a merchant is currently not receiving, concretely: settlement in under a second instead of a receivable that clears later in the week, a single readable record instead of a batch total to reconcile, and a choice between holding dollars in USDC or euros in EURC. On Arc, Circle's Layer 1, gas is paid in USDC and finality lands in under a second, so the settlement asset and the transaction currency are the same thing. @Meridian_Fi Pay accepts USDC and EURC at checkout, gasless for the payer. YieldVault wraps Circle USYC so a merchant balance waiting between sales and payables earns Treasury yield rather than sitting idle, yield from the underlying Treasuries, no fixed or guaranteed rate offered or implied. PaymentStream pays by the second. @FirmataProtocol carries identity, reputation and conditional settlement. Arc testnet, +20 contracts verified on Arcscan. Not mainnet. An independent audit of the contracts and the front end gates migration and it has not happened yet. A merchant trusts a rail with revenue, not with a test balance. The card made the stablecoin spendable everywhere. The next product makes it receivable. Full piece: community.arc.io/home/forum/… Not investment advice #Stablecoins #USDC #Payments

12/ On @arc , Circle's Layer 1, @Meridian_Fi Pay accepts EURC and USDC at checkout, gasless for the payer, settling in under a second on a Sunday as on a Monday. Arc testnet, +20 contracts verified on Arcscan, independent audit before mainnet. €660 million today. Whose euro fills the rest is the real question. Full piece: community.arc.io/home/forum/… Not investment advice #EURC #Stablecoins #MiCA

1/ There is a decision inside every card payment almost no customer has thought about. When a card runs over more than one network, somebody picks which one. That decision moves money. Not the money in the transaction. The money in the fee.

For three years the tokenization argument has mostly been made by people selling it. This month the IMF measured it instead, and the result is less flattering than the pitch and considerably more useful. Chapter 3 of the October Global Financial Stability Report does three things nobody had done together: it sizes the market, it tests the claims empirically, and it names what is actually blocking growth. Start with the size. Tokenized real-world assets excluding repurchase agreements sit at about $65 billion. Tokenized fixed income accounts for $48 billion of that, with credit at $30.4 billion. Tokenized money market funds are near $17.5 billion. Tokenized equities are small, about $2.3 billion, and two platforms hold more than 70% of the value. Then the number everyone will screenshot. Tokenized repo reached $303 billion of average daily volume over a thirty-day window. Two caveats travel with that figure and almost never survive the retelling. It is a volume measure, not an outstanding balance, so it does not compare to the $65 billion. And the Fund notes the activity remains largely concentrated among affiliated entities, which in plain terms means a great deal of it is institutions moving collateral between their own books. Genuinely useful. Not yet a market. Now the part worth reading twice, because it is the first independent test of what tokenized markets deliver rather than what they promise. The Fund analysed the five most liquid tokenized US equities across eleven venues, over two million observations across 365 trading days. Three findings support the case. More than 50% of trading happens outside regular US market hours, so continuous availability is real demand rather than a technical curiosity. About 80% of trades are for less than one share, confirming that fractionalisation answers something. And the overnight on-chain price move is 87% to 99% absorbed by the traditional equity within five minutes of the open, meaning tokenized prices carry information traditional markets act on. Two findings complicate it. Realized volatility is roughly 1.5 times higher. And liquidity is materially worse, particularly on decentralised venues, where a given volume moves the price far more than it would on a conventional exchange. The Fund itself says to read this with caution given how young the market is, and that most of the gap reflects small size rather than bad design. Thin markets are volatile. That is not news. What is new is that an institution with no commercial interest in the answer went and measured it. Then the finding that matters most. The Fund names four constraints on tokenization: legal certainty, regulatory clarity, interoperability, and the availability of safe, sound and scalable settlement assets. Three of those are rules, and rules move at the speed of legislatures. Legal certainty means knowing what rights a token confers and how it is treated in insolvency. Regulatory clarity means applying existing law consistently instead of inventing a parallel regime. Interoperability means assets and money moving between platforms without each network becoming an island. The fourth is different in kind, because it is not waiting on anyone's parliament. The Fund is direct: tokenized trading without a trusted cash leg increases settlement risk, limits delivery versus payment, and prevents end-to-end programmability. And where private settlement assets are used, they bring credit and liquidity risk tied to their issuer. That is why the institutions actually testing this are not testing whether tokens can exist. The Hong Kong Monetary Authority published its Project Ensemble use case report this month, covering tokenized money market fund subscription and redemption, bond settlement, intraday repo for liquidity, cross-border intragroup treasury, tokenized charging station revenue, carbon credits and electronic bills of lading, all settled against tokenized deposits. @Citi , with @The_DTCC and Swift, concluded that banks have invested more than $100 billion in blockchain since 2020 and that what holds back institutional scale is market fragmentation rather than the technology. Swift's shared ledger went live on 9 July with seventeen banks across six continents. And the demand side is more sober than the enthusiasm suggests. Lloyds surveyed 101 financial institution decision makers in April and May. 41% expect tokenization to have a significant or transformational impact, but the weight sits on significant at 30%, with only 11% saying transformational. Asked where the value lands: 60% said payments and settlement speed, 41% collateral and liquidity management, 39% tokenized deposits and cash instruments. They are all testing the same thing. Not whether the asset can be tokenized. Whether money can settle alongside it. On @arc , @circle 's Layer 1, gas is paid in USDC and finality lands in under a second, so the settlement asset and the transaction currency are the same thing. @Meridian_Fi Pay accepts USDC and EURC at checkout, gasless for the payer. YieldVault wraps Circle USYC so a balance waiting on a condition earns Treasury yield rather than sitting idle, yield from the underlying Treasuries, no fixed or guaranteed rate offered or implied. PaymentStream pays by the second. @FirmataProtocol carries identity, reputation and conditional settlement. $17.5 billion of tokenized money market funds is the whole addressable market for holding a waiting balance in Treasury exposure instead of idle cash. Small today. It is the number to watch, because a conditional payment is only economic if the held funds are doing something while they wait. Arc testnet, 20+ contracts verified on Arcscan. Not mainnet, soon. An independent audit of the contracts and the front end gates that migration and it has not happened yet. Arguing that the cash leg must be trustworthy while running unaudited contracts would be the wrong way round. Three constraints are rules. The fourth is the one that moves first. Full piece: community.arc.io/home/forum/… Not investment advice. #Tokenization #Stablecoins #USDC

Soon ..⏳

Global Payments and Worldpay asked 8,027 consumers across 7 markets which jobs they would hand to an AI agent. Censuswide, May 2026. 74% would let it compare prices. 69% would let it research a product. 68% would let it read the reviews. May it complete the purchase on its own? 21%. Another 59% would allow it only with approval on every single purchase. Not a spending limit. Approval, every time. The cliff is not at capability. It is at authorization. Two more numbers that read as a spec: 83% say no one, including the agent, should ever see their real card number. 40% do not know who is accountable when an agent purchase goes wrong. Most of the industry is engineering agents that choose better. Consumers are asking for a checkpoint when money moves. The agents that win will ask for the least autonomy, not the most. community.arc.io/home/forum/… #AgenticAI #Payments #USDC

1/ RedCompass Labs surveyed 300 senior payments professionals across six markets. 12% said their wallet-level AML, KYC and sanctions controls are fully ready for stablecoin payments. 59% said mostly ready.

The @wto published a report on stablecoins and world trade, and the most useful thing in it is a distinction almost nobody in this space draws. Trade in goods needs credit. There is a physical thing moving, a gap between production, shipment and delivery, and somebody has to carry the risk across that gap. So the instrument has to do four jobs at once: lend, control the documents, hold collateral, allocate the risk. Trade in digitally delivered services has none of that. No shipped good, nothing a bank can lend against, no warehouse to inspect. The report's point is that the principal financial requirement there is not credit at all. It is an efficient cross-border payment. Which means a stablecoin is structurally more useful to a software firm invoicing across borders than to a company shipping containers, and the honest version of the pitch tells you which one it is talking to. The report is blunt about the limit, and it is right. "Stablecoins serve purely as a payment layer. They do not supply trade credit, working capital, loans, or guarantees (e.g. letters of credit)." Anyone selling stablecoins into trade should say that out loud before a client finds out. A faster transfer is not a credit facility. Settling in one second says nothing about whether the shipment conformed. Then the same report turns to the instrument that does supply the guarantee, and cites the ICC Banking Commission. @iccwbo Between 65 and 80 percent of documents presented under letters of credit are rejected on first presentation. Read that carefully, because it is easy to overstate. First presentation, not final outcome. Most discrepancies get corrected, and most of those payments complete. Nobody is claiming two thirds of trade finance collapses. What the number measures is friction, and the cause is a rule, not incompetence. Letters of credit run on strict compliance: every document has to be accurate, complete and consistent with the terms all parties agreed. The WTO's phrasing is that even tiny mismatches can halt payment. A misspelled port name. A date one day outside the window. Goods described with one word that differs from the contract. So the instrument built to certify that a condition was met cannot, two times out of three, certify it on first reading. It gets there through correspondence, corrected documents, re-presentation and weeks. Then the part that should bother everyone more than either of those. Across emerging markets surveyed in joint WTO and International Finance Corporation studies covering West Africa, the Mekong region and Central America, only 2 to 20 percent of trade flows are supported by trade finance at all. In Mexico, roughly a quarter of importers and exporters have any access to credit. The largest share of traders is unserved: self-financing from working capital, borrowing informally, paying cash in advance, or shipping on open account and carrying the risk themselves. Cash in advance means the buyer funds the seller's production and wears the risk that nothing compliant arrives. Open account means the seller ships and wears the risk of not being paid. Neither is a choice between instruments. Both are the absence of one. For scale, the WTO cites a @BIS_org estimate of 6.5 to 8 trillion dollars of bank-intermediated trade finance, plus 2.5 to 3 trillion in non-bank forms such as credit insurance and factoring. That puts the total between 9 and 11 trillion, against 19 trillion of total merchandise trade in the same year. Read it as old: those are 2011 figures, published by the Bank for International Settlements in January 2014. Now the finding that cuts against my own argument, and it is better to quote it than to wait for someone to quote it at you. Researchers at @bancaditalia ran 200 USDC transfers across ten corridors linking Italy with Argentina, Brazil, South Africa, the United Arab Emirates and Japan. Total cost ranged from 0.3 percent to almost 9 percent of the amount transferred, driven primarily by the fiat on-ramp and off-ramp, not by the on-chain transfer. The WTO's own conclusion is the sober one. A payment is only as efficient as its slowest, most expensive component. Swift's data says the same thing from inside the incumbent system. Roughly 80 percent of a cross-border payment's journey time happens in the last mile, after the funds reach the beneficiary bank. The international leg accounts for under 20 percent, and three quarters of payments on its network reach the beneficiary bank within ten minutes. The international leg is not the problem any more, in either system. The ramps and the last mile are. So what does conditional settlement actually change? Less than the pitch usually claims, and I would rather say that first. It does not supply credit. It does not underwrite performance. It does not extend working capital. It does not replace a guarantee from a bank with a balance sheet. Anyone telling you a smart contract is a letter of credit is describing something else. What it changes is when value releases and on what evidence. Under ERC-8183, funds are committed against a condition, released when that condition is evidenced, returned when it is not. Under ERC-8004, the counterparty carries an identity and a reputation record instead of being an address with no history. Settled over x402 on @arc , where gas is USDC and finality lands in under a second, so the whole thing stays in the currency of the transaction. A stablecoin transfer answers when the money moved. A letter of credit answers whether the documents matched, eventually. Neither produces a machine-checkable record, available to both sides at the moment of dispute, of what was promised and what was evidenced. There is a reason this only works as one system rather than a bolt-on, and the paperwork problem is where you can see it. Funds held against a condition are funds not earning, which is exactly the trapped working capital that makes cash in advance expensive for the buyer. @Meridian_Fi Pay accepts USDC and EURC at checkout, gasless for the payer, settling in under a second. YieldVault wraps @circle USYC so a balance waiting on a condition earns Treasury yield rather than sitting idle, withdrawable at any time, yield from the underlying Treasuries, no fixed or guaranteed rate offered or implied. PaymentStream pays by the second for work delivered continuously rather than monthly. A balance can sit in USDC, EURC, USYC or cirBTC and move between them without leaving the chain. @FirmataProtocol carries the identity, reputation and conditional settlement layer. Putting the escrow and the yield-bearing asset on the same ledger, redeemable into one another in near real time, is what stops a conditional payment from costing the buyer the float. Firmata does not perform AML or KYC and does not replace anybody's obligations. A broker keeps its own customer due diligence exactly as it does today. What changes is that the settlement carries a proof rather than a promise. All of it runs on Arc testnet. + Twenty Five contracts, each verified on Arcscan from the second of deploy, roughly six hundred thousand transactions and four thousand on-chain holders. Not mainnet. An independent security audit covering the contracts and the front end gates that migration, and it has not happened yet. For a piece about trade finance that matters more than usual. The entire value of a guarantee is that somebody credible stands behind it. Claiming to improve on a letter of credit while running unaudited contracts would be the wrong way round. Full piece: community.arc.io/home/forum/… Not investment advice. #TradeFinance #Stablecoins #USDC #EURC

1/ Six banks published a joint paper on agentic commerce this month. ASB, Bank of America, Capital One, Commonwealth Bank, ING, NatWest. Two globally systemically important. Voluntary, non-binding, endorses nobody. And it reads like a purchase order.

A consultation asks what you think. A test rig finds out what breaks. The @TheFCA 's Smart Data Accelerator did not open a consultation on AI agents. It commissioned architecture research from Raidiam and Oxford's Saïd Business School, led independently by Prof Pinar Ozcan. The output is blueprints, tested in the regulator's own Digital Sandbox against synthetic data. Four dimensions. The second one is the tell: Technology, covering agentic AI, distributed ledger technology, and quantum-readiness. Agentic AI is not in a futures paper. It sits inside an infrastructure brief, next to quantum-readiness. Quantum is on that list because cryptography has a replacement cycle you cannot start late. Putting agents beside it says the same thing. Three architectures being compared: centrally coordinated, federated, hybrid. The federated column already has running code. Nobody writing agent standards had a central coordinator available, so they specified the hard case first. And it settles somewhere. @arc 's mainnet opened 16 September. @circle 's L1, USDC as the native gas token, so an agent never holds a second volatile asset just to pay a fee. 99.3% of observed x402 value settles in USDC. When software pays software today, it pays in Circle's dollar. Still testnet for us. Audit before mainnet. @Meridian_Fi @FirmataProtocol community.arc.io/home/forum/… #OpenFinance #AgenticAI #USDC

Same research. Same table. Both sides. 72% of bankers expect stablecoins to improve efficiency and protect margins. 75% of their clients expect to pay less per transaction. They cannot both be right. It is the same saving. @IBM Institute for Business Value, 500 corporate CFOs and treasurers, 250 banking executives, with BAFT and BIAN. Published by the authors ahead of the formal launch on 27 September. The tiebreaker is not in that pair. It is in the next one. 57% of those finance leaders are moving the control point away from their bank. 33% want their own agentic AI routing payments, liquidity and treasury in near real time. 24% want multi-bank platforms as the interface. An orchestration layer that compares providers on price and executes continuously is a device for making a margin visible and routing around it. Meanwhile under 20% of the bankers name revenue as the primary outcome of their AI spend. Automation and cost reduction lead. One side automates what it already does. The other decides who it does business with. Then a separate survey, different population. Bank Director, 15 September, US bank executives and directors: 63% say stablecoins are overhyped. 10% have a stablecoin strategy. 72% expect to keep the margin. 10% have a plan to. Capability is not the alibi. Kinexys is past $4T since inception, ~$7bn a day. Both surveys are sponsored and both measure intent, not behaviour. Said up front. Come to @arc House to read the full post, link in the comment 👇 #CorporateBanking #Stablecoins #AgenticAI

trust and verifiability for agents was a niche conversation a year ago. now it's the headline 🫡 the teams who spent that year building instead of waiting are about to find out it was worth it @FirmataProtocol ⏳

Soon, @arc here we are 👀 Days and nights of work... and that’s been going on since November 1, 2025! @Meridian_Fi @FirmataProtocol ⏳

1/ Most chains launch as an empty building, then spend three years persuading people to move the furniture in. @arc launched 16 September. By 25 September it had cash, collateral, a credit line, a money market fund and a live FX venue. Nine days.

On 14 Sept the FCA and Bank of England published FS26/1 on tokenisation in UK wholesale markets. 123 responses, including BlackRock, BNY, Euroclear, Nasdaq, Coinbase, Ripple, Chainlink, Tether. Everyone will cover the collateral finding. The paragraph worth keeping isn't about tokenisation. One of the regulatory principles: all regulated activities must have an accountable, regulated person responsible for regulatory compliance. Respondents pushed back, arguing it would be hard for DeFi firms to meet and so wasn't really tech neutral. The answer was no. Regulated activities require an identifiable person responsible for that activity, and the FCA sees no case to change the perimeter. Its reason is one line: it cannot accept solutions where customers have no recourse when assets are lost or stolen. Recourse needs somebody to have recourse against. Now the IMF, April, Note 2026/004, on a completely unrelated market: "Most payment regimes require that a payment order be traceable to an authorized instruction from an account holder or its legally recognized agent. Agent-initiated payments challenge this model, as individual transactions may not correspond to explicit, transaction-level instructions." Authorization becomes structural and mandate based. Two authorities, two unrelated markets, five months apart, same requirement, and neither has a mechanism for it. Why that matters: you told your bank to pay a supplier, one instruction, one payment, one name. You told an agent what you wanted, four hundred payments, every one authorised, none of them instructed. The legal question has an answer at the mandate and no answer at the transaction. The IMF's risk table has three columns. Source, who bears the cost, and whether there's a market failure justifying policy intervention. Eleven risks, eleven yeses. For the instruction gap the cost lands on account holders, on payment providers through disputes, and on payment systems through operational strain. Two of those three never made a decision about the agent. That's an externality, which is why the third column says yes. Separate two things that keep getting merged. Attribution asks who is responsible for this payment. Performance asks whether the thing paid for happened. The IMF's authorization layer: AP2 mandates, OAuth, ERC-1812 claims, ERC-780 registries, ERC-6900 spend limits, and ERC-8004 for agent identity and reputation, named in an IMF table under "know your agent". All of it attribution and permission. Nothing in that layer holds funds against a condition. One layer down the principle is already settled, and it took three years. @circle applied to the OCC in June 2025 and got final approval on 10 July 2026, USDC reserves in cash and short-dated Treasuries attested monthly, EURC authorised under MiCA, listed on the NYSE. A supervisor asking who is accountable for the instrument has an address. Asking who is accountable for the agent's payment, it doesn't. #AgenticAI #Payments #Stablecoins