Bankers have a name for the risk of paying your half of a trade and never receiving the other half. They call it Herstatt risk, named after a German bank called Herstatt that collapsed in 1974.

On 26 June that year, several banks sold Deutsche Marks to Herstatt, a mid-sized bank in Cologne. Herstatt owed them US dollars in return.

The banks paid their marks to Herstatt in Germany during the day. Herstatt would pay the dollars in New York, where the banking day runs hours behind Germany's.

At the end of the German banking day, regulators shut Herstatt down. The banks had already paid their marks. The dollars, due later in New York, never came.

Their trades had been agreed and confirmed, and none of that helped. All they held was a claim on a closed bank.

A whole layer of financial plumbing now exists so that day does not happen again. eThe collapse also fixed what a treasurer means by "settled": the moment value has moved and the law will not let anyone take it back, even in an insolvency.

Fifty-two years later, Circle, the company behind the USDC stablecoin, makes a different promise. Its new blockchain, Arc, went live in September, and Circle's introductory video says transactions on it "are either pending or fully final. There is no probabilistic waiting period." That model, the video adds, "aligns with the expectations of traditional capital markets and regulatory settlement standards." Circle says Arc confirms a transaction as final in about 350 milliseconds (roughly a blink).

Circle's own disclosures add that Arc has not been reviewed or approved by any regulator.

When a stablecoin payment is final on the chain, is it settled in the way a treasurer means?

What does the chain leave unanswered?

When I started this piece in July, I thought the argument was about speed. Ethereum, the biggest home for stablecoins, records a new batch of transactions every twelve seconds but only treats them as irreversible after about thirteen to sixteen minutes.

Newer blockchains built for payments do it far faster. Arc, Stripe's Tempo and Plasma all say a transaction is final, with no chance of reversal, in under a second. They manage it by limiting the validators, the companies that confirm transactions, to a known, approved group. Arc's include BlackRock, Visa, Mastercard and DTCC, which settles almost every US share trade. Most stablecoins still sit on Ethereum and Tron, where finality takes from about a minute to a quarter of an hour, but the direction is obvious.

Ethereum is heading the same way without limiting who can run it: the Ethereum Foundation's long-range plan, published in February, aims to cut finality to as little as eight seconds by the end of the decade.

So speed has stopped being the interesting part. A transfer that cannot be reversed still leaves three questions open:

1. Has the recipient been paid
2. Can they swap what arrived for its full face value (what bankers call par)

3. Are they protected if a counterparty fails? The chain answers none of them.

Four stacked rows for one stablecoin transfer: final on the chain (the transfer cannot be reversed), then three open questions: paid (decided by contract and commercial law), worth a full dollar (a claim on the issuer, which can freeze tokens), and protected if a counterparty goes bust (only inside officially designated systems).

Starting with whether the recipient has been paid as that is a matter of contract and commercial law.

Two companies can agree that a stablecoin transfer settles an invoice, and no regulator has to bless it. What US law lacks is a rule for when they have not agreed. Three law professors, in a forthcoming Yale Law Journal article, call the GENIUS Act, the 2025 US stablecoin law, "conspicuously silent" on whether paying in stablecoin settles a debt. They point out there is no stablecoin equivalent of the rules that make a wire transfer final.

The second question is what the recipient now holds. Hyun Song Shin put it plainly at a European Central Bank forum in July: "if I send a stablecoin to someone, the receiver acquires a claim on the issuer." Shin ran the economics department of the Bank for International Settlements, the central banks' own bank, before becoming Governor of the Bank of Korea in April.

A bank payment ends a bit differently. The deposits move and the banks settle with each other through the central bank, so a dollar at one bank is accepted as a dollar at any other. In Shin's words, "the singleness of money is guaranteed by design."

A claim on an issuer comes with the issuer's power to freeze it. On 29 October, Binance said hackers had taken about $6.3bn. Within hours, Tether and Circle, had frozen more than $2.3bn of the stolen stablecoins.

The freeze did not reverse anything. The transfers stayed final. The tokens stayed in the addresses they were sent to, but the issuers had made them impossible to move. A transfer can be final and still deliver something the issuer can freeze.

Few will mourn the thief, and from January the GENIUS Act will require regulated US issuers to be able to comply with lawful orders to freeze tokens.

The third question is the narrowest and matters most in a crisis: if a counterparty collapses tomorrow, can an administrator undo the payment? BNY's Victor O'Laughlen gave the client's version at a stablecoin conference in New York in June: "are you going to be tied up in an FTX workout for three years? That's not why clients use BNY triparty."

Why do banks wait before they settle?

Because waiting lets them net. Before any money moves, banks work out who owes what and cancel out what they can: if I owe you 100 and you owe me 80, only 20 needs to move. Bankers call the working out "clearing" and the money moving "settlement".

Netting is what makes the system affordable. Frank La Salla, DTCC's chief executive, said at a conference in New York in late October that DTCC settles $20 trillion of trades a day by netting them down first. Paying every trade in full, one at a time, is called settling gross, and as he put it, "there is not enough cash in the universe to settle gross. There isn't."

The wait has a price though, and Herstatt's counterparties paid it: they had handed over their marks before the dollars arrived. Banking's answer was to keep the delay and make it safer. In 2002 the biggest banks launched CLS, which pays out both halves of a currency trade at the same moment or not at all, while still netting what each bank owes. It settled an average of $8.11 trillion a day in 2025, across eighteen currencies, and it has legal protection that a trade on a public blockchain does not.

A blockchain can remove the delay altogether. A trade and its payment happen in the same step, and either both go through or neither does. Engineers call this atomic settlement. It is new, and worth being impressed by.

La Salla's other point: "by its very nature, blockchain settles gross. We settle net." Trades can be netted before they reach the blockchain, but then you are waiting again, even if only for minutes. So the real choice is how long to wait and who carries the risk while you do. That is a decision for whoever runs the system.

Who stands behind "final"?

The clearest live answer arrived on 28 October, when DTCC switched on its Tokenization Service, which lets firms hold digital tokens of certain shares, funds and US Treasuries that DTCC keeps for them. Twenty firms have signed up, including JPMorgan, Citi and Morgan Stanley. The tokens run on two blockchains, one public and one DTCC runs itself. In July I wrote about the first live trades, when JPMorgan used tokenised shares of an Invesco fund to meet a margin call at CME.

Under the SEC letter that cleared the service, DTCC's own records, kept off the blockchain, remain the official record of who owns what.

In DTCC's risk checks the tokens count for nothing, and DTCC can change transactions or destroy tokens after a mistaken or illegal transfer. The cash still moves through ordinary bank accounts; moving it on the blockchain is a 2027 plan.

DTCC is one of Arc's founding validators, and it still keeps the legal record off the chain. Nadine Chakar, who runs DTCC Digital Assets, put it simply in August: "We are the official recordkeeper."

A court can still treat a blockchain transfer as complete, or a debt as paid, without any special law. Settlement-finality laws, such as the EU's Settlement Finality Directive, do something narrower. They guarantee that a payment made through a protected system cannot be undone if one of the parties goes bust.

That protection only goes to systems a regulator has officially named, or "designated": set-ups with common rules, members the regulator can see and an operator who answers for the whole thing. A permissionless blockchain, which anyone can help run, has no such operator. So every route to this protection on a blockchain so far runs through a layer with a name on it: a depository's books, a central bank's settlement system. (October's piece on onchain equity covered the related question of who keeps the register.)

Three stacked layers: the blockchain, an operator with a rulebook (DTCC, ECB Pontes), and settlement-finality law. Protection if a counterparty goes bust attaches to the operator layer; an open blockchain anyone can help run has no operator to designate. Courts can still recognise a transfer.

Who turns a stablecoin back into cash?

Tony McLaughlin spent two decades at Citi, where he designed one of the first blueprints for putting bank money on a shared ledger, before leaving in 2025 to start Ubyx. He puts the problem in one line: "there is a very big difference between trading and clearing."

Trading is selling your stablecoin on an exchange for whatever buyers will pay. Clearing is handing it to your own bank and getting a full dollar back from the issuer. "You cannot take a stablecoin to your bank and get par value," he said on a podcast in September, which by his measure makes stablecoins "inferior to checks."

Ubyx wants to be the clearing house that fixes that. It says it made its first live redemption at full value in October 2025, and it has since taken investment from Barclays and added BitGo's trust bank as a settlement agent. It is still early: there are no published volumes, and I could find only one named receiving institution.

Banks are already questioning whether stablecoins count as cash. At the ECB's money market conference on 6 November, Nate Wuerffel, BNY's global head of market structure, talked about using a stablecoin as the cash in a repo, a short-term loan secured on a government bond. That, he said, "is really a collateral for collateral transaction, it's not cash for collateral... you might need to haircut both sides."

I’m assuming a haircut means valuing something at less than face value to cover the risk of holding it.

Where is the law going?

Slowly, and in Europe first.

The European Commission proposed in December 2025 to turn the Settlement Finality Directive into a regulation and to let blockchain-based systems be designated. A designated blockchain system would have to "implement mechanisms guaranteeing deterministic and legally enforceable finality moments": a fixed point, recognised in law, after which a payment cannot be undone. Technical standards would then set out how that point could be defined for "probabilistic or layered finality models", chains like today's Ethereum, where certainty builds up over time. Even those, the text says, may reach legal certainty "depending on the structure and rules of the system."

On current projections, the rules would not apply until around 2029.

The ECB has not waited for the law, and its own project shows the same split. Pontes, which links blockchain platforms to central bank money in euros, went live on 21 September. A cash token can move on the blockchain, but the payment only becomes final when the matching payment goes through T2, the ECB's main payment system. Piero Cipollone, the ECB board member who leads the work, lists "offering immediate settlement finality" on the central banks' own blockchain as a future upgrade.

The US has gone the other way. The GENIUS Act takes effect on 18 January 2027 and gives holders priority over an issuer's reserves in insolvency. That protects the value of a stablecoin. It does not say whether a transfer is final or a debt is paid.

The market-structure bill that might have gone further, CLARITY, failed a Senate procedural vote 49 to 50 on 15 September, and Democrats won both chambers on 3 November. A new federal settlement-finality law now looks unlikely for at least two years.

Does any of this need solving?

The strongest objection is that none of this has to be solved. Stablecoins can stay a fast side network, swapped back into dollars when needed, while institutions keep settling big obligations where the law already protects them. On that view the clearing layer is a business waiting for volume that may never arrive.

The optimist's version comes from Circle's chief executive, Jeremy Allaire, who told Sibos, Swift's annual banking conference, in October that once GENIUS takes effect, digital dollars like USDC "will be able to be used by banks, by capital markets firms, by exchanges, in custody and clearing, in the guts of the financial system." If he is right, regulated issuers will answer most of these questions without anyone redesigning settlement law.

Twenty firms signing up to DTCC's service is a real signal for securities and collateral. For payments, even banks exploring stablecoins as cash are questioning whether to treat them at full value.

My prediction, by the end of 2028, the big banks that accept stablecoins at full value will do it through a clearing layer that someone runs and a regulator supervises, not by trusting the blockchain's finality on its own.

If I am wrong, it will be because a major bank starts taking stablecoins straight from customers' wallets at full value and carries the issuer risk itself. On the legal side, watch Brussels: if the EU's technical standards let a designated system take a permissionless chain's own finality as the legal moment, the operator matters less than I think.

See you next week.

James Smith

Many of my articles start with a reader's question, from people at banks and central banks. If you're working on any of this, I'd like to hear from you. Message me on LinkedIn or Telegram (@snapcrackle).