Your OTC desk agrees a price today and settles tomorrow, and everyone sleeps fine. Not because the overnight risk disappeared. It never disappears. Someone is always holding it - and in traditional finance, the holder has a name.
Between agreement and settlement, the risk that your counterparty fails to deliver is carried by a very specific stack: a credit department that decided how much exposure to allow you, an ISDA master agreement that defines what happens if you default, collateral posted against the position, and, at the bottom of everything, a court system that can seize assets from a legal entity. T+1 is not safe because nothing can go wrong overnight. It is safe because when something goes wrong, there is a person to sue.
What the gap costs when the holder fails
Finance learned the price of that gap in September 2008. When Lehman Brothers filed for bankruptcy, it was a counterparty on roughly 900,000 open derivatives trades - forwards, swaps, options, all instruments that live in the space between agreement and settlement. Every one of those trades had Lehman's balance sheet as the thing holding the gap. Unwinding the book took more than a decade of litigation across multiple jurisdictions.
The regulatory response was not to close the gap. It was to give the gap a stronger custodian. The G20 pushed standardized OTC derivatives into central clearinghouses, mandated initial and variation margin, and in the process made CCPs some of the most systemically important institutions on the planet. It broadly works. But notice what every layer of the solution assumes: counterparties that have legal names, sign master agreements, maintain margin accounts, and can be pursued through courts when they fail.
The counterparty that can't be sued
Now put an autonomous agent on one side of that forward.
An agent has a wallet and a signing key. It does not have a legal name. It cannot sign an ISDA master agreement. It cannot open an account at a clearinghouse, because clearinghouse onboarding is built for regulated entities with balance sheets and compliance officers. And if it defaults, there is nothing to sue - a fresh wallet has no assets to seize and no reputation it cannot abandon.
Every tool traditional finance uses to hold overnight risk assumes legal personhood. Agents have none. Which explains the shape of today's agent-payment landscape: almost every rail an agent can reach settles on the spot, in the next block or two. Spot swap, spot payment, spot escrow. "Settle now" is the only thing the stack can say safely, because "settle later" requires someone to hold the gap - and there is no one.
That is a real expressiveness loss. Forwards exist because "now" and "later" are genuinely different needs: hedging a known future obligation, matching a payment schedule, locking a price ahead of expected volatility. An agent economy that can only settle on the spot cannot express one of the oldest instruments in finance.
Holding the gap with code instead of a balance sheet
There is a third option between "trust a balance sheet" and "spot only": make the settlement itself atomic at the forward date.
The mechanism is one added constraint on a hash-time-locked contract. A standard HTLC already has two knobs: a hashlock (the claim requires revealing a secret) and a timelock (if nothing happens, both sides refund). A forward adds a third: the moment the claim becomes valid. Both parties lock their legs at agreement time, but the withdraw path only opens inside a window - from the agreed settlement date until the timeout. Before the window, neither side can pull the trade forward. Inside it, settlement is atomic: one claim reveals the secret that unlocks the other leg. After it, if settlement never happened, both sides refund.
Walk through what happened to the overnight risk. Between agreement and the settlement window, neither party is exposed to the other's balance sheet, because there is no balance sheet in the middle - both legs sit locked in contracts with deterministic refund paths. Your money never leaves your wallet's control in the sense that matters: no custodian holds it, and the worst-case outcome is your own funds coming back after the timeout. Default, in the traditional sense of "my counterparty owes me delivery and cannot pay," stops being an event category. It is not forbidden. It is unexpressible.
The honest ledger
This design pays for that property, and it is worth being precise about the price.
Full collateralization. A cleared forward in traditional finance is margin-efficient - you post a fraction of notional. An atomic forward locks the full notional on both sides for the entire gap. That is real opportunity cost, and for leveraged trading desks it may be the wrong trade. For an agent that would otherwise have no safe way to express "later" at all, the comparison is not against 5% margin - it is against nothing.
Market risk stays. A forward fixes the price by design. If the market moves, one side will be settling at a worse-than-market rate. The mechanism removes counterparty risk, not price risk - exactly like any forward.
The abandonment option. The side holding the secret can choose silence: reveal nothing, let the window close, and both legs refund. The disadvantaged party never loses principal - but their capital was locked for the duration, for nothing. This is the classic optionality critique of HTLCs, and it is fair. It is also why a settlement layer needs an economic layer on top: completion bonds that an agent stakes on its commitment, forfeited to the counterparty it leaves waiting. We wrote up that mechanism - execution rewards - yesterday. The residual risk in an atomic forward is time, not principal, and time is exactly the thing a bond can price.
Where this sits
None of this competes with the payment rails being built for agents right now - HTTP-native micropayments, facilitator APIs, per-request billing. Those move a single leg for a completed service call, and they are good at it. A forward is a different object: two legs, two chains, a deliberate gap in between. That is settlement-layer work, underneath the payment layer.
At Hashlock we are building this as part of a sealed-bid RFQ protocol fused with HTLC settlement, exposed to agents as MCP tools (the server is @hashlock-tech/mcp on npm, currently 0.5.1). Status, stated precisely: the settlement core is live end-to-end on Ethereum mainnet; Sui contracts are deployed and CLI-tested, not yet live; Bitcoin is validated on signet, mainnet pending. The forward window design above is one of the five primitives we are building on that core - the mechanics are in our docs: https://hashlock.markets/docs?utm_source=devto&utm_medium=article&utm_campaign=2026-08-04-forward-otc-overnight-risk
The protocol design is written up in more depth in our whitepaper on SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6712722
The question
Two audiences, one question each.
If you run size through an OTC desk: would you accept locking full notional for 24 hours to delete counterparty risk entirely, or is margin efficiency the whole game - and someone to sue an acceptable substitute for certainty?
If you build agents: what is the longest gap between "agreed" and "settled" your stack can express today - and what trade did you not do because it couldn't?
Top comments (0)