Someone Has to Authorize the Trade.

Last week I told you I was at the wrong end of the curve.

Seven editions building toward the household, the boomer parent, the spending ceiling set and forgotten, the receipt someone has to sign. In #017 I turned the argument around: infrastructure never diffuses laggards-first. The first adopters of governed agents are not the least technically fluent. They are the most. Crypto-native. Already algorithmic. Already transacting at machine speed on mandates set by a human who left the building.

That was the pivot. This week I want to go concrete on the mechanism.

Because the question I left unanswered in #017 is the one the audit committee, the prime broker, and the regulator will all ask at the exact same moment:

Who told the agent to make that trade?

The Trade

Picture the position every one of these compute-futures products points toward.

In early July, ICE and NATIVX announced plans to launch cash-settled GPU compute futures tied to the COIL Index, an energy-normalized benchmark tracking the price of compute and connectivity. It is the latest product in a wave that arrived across 2026: Ornn's compute price index went live on the Bloomberg Terminal in April; ICE and Ornn announced their own GPU compute futures that same spring; CME filed with Silicon Data for a competing contract. The benchmark race is not settled. But the direction is.

Compute has a bid and an ask. It is getting a futures curve. And underneath all of this: public Bitcoin miners have signed more than $70 billion in AI and HPC contracts, converting their power assets and shells into compute supply. The physical layer is being built. The paper market on top of it is arriving.

So picture the trade.

Leg one: An agent hedges the compute position. The firm has forward exposure to spot GPU capacity, committed neocloud spend, inference load that tracks revenue, a data center buildout that depends on a stable cost basis. The agent places a futures position on a COIL-indexed contract to offset that exposure. This leg exists now. The contracts are coming.

Leg two: That paper position needs to convert to actual capacity. An Exchange-for-Physical transaction, the agent exits the futures leg and executes a matched spot purchase from a neocloud or capacity marketplace, priced against the benchmark. The cash-settled futures are live; the physical-conversion layer is where this arc is heading. Every HPC and GPU marketplace in the stack, CoreWeave, VoltagePark, Vast.ai, Akash, is a potential delivery counterparty.

Leg three: Settlement finalizes on-chain. Increasingly, the plumbing for this runs on stablecoin rails. Programmable, machine-speed, no human intermediary at execution time.

Three legs. Machine speed. From hedge to physical delivery to cryptographic settlement, the whole transaction might run in minutes.

Now ask the question.

The Question Nobody Has Built the Answer To

Who authorized this?

Not in a vague, philosophical sense. In the operational, compliance-grade sense that every institution touching regulated markets has to answer.

The agent placed the futures position. At what size? The firm's hedging mandate presumably says something about that, but does it say it in a form the counterparty can verify? Does the exchange know whether this agent had authorization to commit to that notional, or is it just trusting the API key?

The agent executed the spot buy. Against which counterparties was it cleared to transact? What was the price tolerance? Who set the ceiling? Can you prove, after the fact, that the execution stayed inside the mandate, that the agent did not exceed scope, did not hit a counterparty the CRO had blacklisted, did not blow through the spend limit by a rounding error at machine speed?

The settlement hit the chain. There is now a cryptographic record of the transaction. But the cryptographic record of authorization, the proof that a human, with a mandate, within a defined scope, told the agent to do this, that record does not exist unless someone built it first.

The audit committee will ask. The prime broker runs counterparty risk on the answer. If the CFTC approves the CME/Silicon Data filing and compute futures become a regulated instrument, position-limit reporting and anti-manipulation obligations will build a formal documentation requirement around it.

Three legs. Every one of them is an authorization event. None of them currently has a portable proof layer.

The Architecture That Closes the Gap

Here is what "attributable authorization" means in practice.

Before the agent places the hedge: a scoped credential. This agent, acting under this mandate, is authorized to place compute-futures positions up to this notional, against these cleared counterparties, inside this window. The credential is not an API key. It is not a role permission in an internal system. It is a structured, cryptographically signed delegation from the principal, the human or committee that set the mandate, that travels with the agent into the transaction.

When the agent executes: the credential is presented to the counterparty. Not as a claim the agent makes about itself. As a verifiable attestation: here is the scope I was given, here is who gave it, here is the window it is valid for. The neocloud on the spot-buy leg can verify authorization before settlement. The prime broker can verify before clearing.

After the trade: a receipt exists that was not constructed retrospectively. The credential that authorized the transaction and the transaction record that consumed it are linked. The CFO, the CRO, the regulator, the external auditor, they all pull the same artifact. It says: this is what was authorized, this is what was executed, this is the delta, this is who was responsible.

That architecture is not exotic. It is what every regulated market demands of its human participants. Position limits, authorized-trader designations, know-your-customer documentation, these exist because "who authorized this" is a non-negotiable question in any market that is large enough to cause systemic harm when something goes wrong.

The agents arriving in compute markets are not exempt from that question. They are subject to a faster version of it.

Accords governs what agents do on the open internet.

That line is not metaphor at this point. Compute futures settle on crypto rails: which is not a futurist projection, it is the settlement architecture the EFP layer is being built around. Crypto rails are the open internet. No platform owns the surface. No intermediary sits between the agent and the counterparty to serve as the implicit trust anchor.

If the authorization layer does not travel with the agent, it does not exist at the moment the trade settles. There is no fallback.

AccordsVault is the answer to that architecture. The scoped credential. The cryptographic receipt. The proof layer that connects the mandate that was set, by the quant desk principal, the CFO, the investment committee, to the transaction that was executed, at machine speed, in three legs, on a market that is becoming regulated faster than anyone expected.

The household principal who set a spending ceiling and left the room is, structurally, the same problem. A human set a mandate. An agent consumed it. The question is whether there is a proof artifact linking the two. The household version is slower, softer, and years away from being the frontier. The compute version is this quarter.

Falsifiable Close

Here is what breaks this thesis.

If the CME/Silicon Data futures filing stalls at the CFTC, and the ICE x Ornn listing stays in regulatory limbo past Q3, and the ICE x NATIVX COIL contracts similarly cannot clear approval, then the financialization of compute is moving slower than this arc assumes. The machine-speed agent trader who needs a portable authorization layer is not yet executing in volume that generates compliance pressure. The governance gap is real but the urgency is not.

Watch two things: the CFTC clock on the CME/Silicon Data contract (the most scrutinized of the three, given CME's scale); and whether ICE lists either of its compute futures products, COIL or Ornn-indexed, with real market-maker participation before year-end.

If the volume is there, the governance obligation follows. Not because someone decided it was a nice idea, but because regulated markets do not allow "the agent decided" as a complete answer to "who authorized this trade."

The next edition goes one level deeper: the EFP seam. The exact moment where a paper hedge becomes a physical delivery is the highest-value governance surface in the entire stack. That is where the authorization layer earns its keep.

Max