The most consequential line to cross a trading terminal this month did not originate from a protocol upgrade or a central bank. It came from the US Treasury Secretary, who characterized the risk warnings issued by leading AI laboratories as "alarmist" โ and in doing so quietly redrew the boundary between what the state will police and what the market must police for itself. For anyone who has spent a decade watching crypto's regulatory pendulum swing, the cadence was familiar. It was the same argument, in a different costume, that the industry heard when it was told that accountability and self-regulation would substitute for enforcement. The question worth asking is not whether the warning was overstated. It is what fills the vacuum once the regulator decides the warning was noise.
The backdrop deserves more weight than the quotation itself. The current US administration has moved decisively toward an innovation-first posture on artificial intelligence, unwinding the executive order that had previously tethered federal agencies to precautionary risk assessments and replacing it with a framework that treats deployment speed as a national asset. The Treasury Secretary's remark lands inside that architecture, not outside it. When a fiscal authority โ not a technology regulator โ labels safety warnings alarmist, the message is aimed less at researchers than at capital markets: the compliance discount on AI equities should narrow, the litigation overhang should be repriced, and the frontier labs should be free to ship.
That framing collides with Europe, where the AI Act has institutionalized the opposite instinct, treating ex-ante risk classification as a precondition for market access. The divergence is not academic. It creates a regulatory arbitrage that will pull compute, talent, and incorporation toward whichever jurisdiction prices risk most cheaply. What the brief coverage of this episode omitted is the detail that would let us weigh it: the venue, the full transcript, the named targets, any evidence supporting the claim that warnings were exaggerated. A policy signal delivered without evidence is still a signal โ it simply tells us about intent rather than about facts.
Here is where the crypto lens earns its keep. The debate has been framed as a binary between precaution and innovation, as though withdrawing external accountability reduces the total demand for trust. It does not. It relocates that demand into the private sector, and the private sector's only scalable answer to unverifiable claims is cryptographic proof. Deregulation does not eliminate the need for verification; it privatizes it and hands the bill to the market.
I spent the better part of last year modeling exactly this. The premise of verifiable compute markets is straightforward: if an AI system produces an output that carries economic consequence โ a credit decision, a medical triage, a settlement instruction โ the party relying on that output needs a way to confirm that a specific model, running specific weights, produced it. In an environment where a regulator certifies the model, that confirmation is a stamp on a document. In an environment where no regulator does, the confirmation has to be generated by the system itself.
The mechanisms are less mature than the pitch decks suggest. Trusted execution environments offer the cheapest path, but they anchor security in silicon vendors rather than mathematics, which merely moves the trust assumption one layer down. Zero-knowledge machine learning offers cryptographic guarantees, but the proving overhead for anything beyond small models remains punishing โ the cost of proving a transformer inference can exceed the cost of computing it by orders of magnitude. Attestation layers and provenance registries are more practical, yet they inherit the oldest problem in the oracle problem: the proof is only as honest as the feed. None of this is disqualifying. It is simply the difference between a market that exists and a market that has been narrated into existence.
A cross-border payment researcher sees this earlier than most. When I was mapping correspondent banking flows, the binding constraint was never the technology โ it was the reconciliation of trust across jurisdictions that do not share legal recourse. SWIFT works because a network of regulated institutions agrees to be accountable to one another. Strip out the regulator and you have to rebuild that accountability from cryptography. That is precisely the trade the AI industry is now being invited to make, and it has not yet built the equivalent of SWIFT.
The parallel to DeFi is uncomfortable and precise. DeFi's glass house shatters under its own weight not because the code failed, but because the trust assumptions buried beneath the code were never audited with the same rigor as the code itself. Audits became marketing. Decentralization became a slide. And when the flow stopped, the industry discovered that the collateral it had pledged was priced by the same reflexivity it claimed to escape. AI's verification layer is now walking the same road. Every week brings a new attestation protocol promising to make model outputs provable, and a meaningful fraction of them are packaging a TEE wrapper and calling it trustlessness.
Consider the numbers that would settle the argument. My own modeling put the addressable market for verifiable data sources at roughly half a billion dollars by 2028 โ a figure that assumed regulatory pressure would sustain demand. Remove that pressure and the figure does not collapse; it redistributes, from compliance-driven buyers to liability-driven ones. The total is similar. The composition is entirely different, and composition is what determines which protocols survive.
There is a sharper read on the Treasury Secretary's remark, and it concerns competitive structure. The safety narrative advanced by frontier labs is not only a public good; it is a barrier to entry. A regime that requires pre-deployment safety testing, licensing, or mandatory evaluation favors incumbents who already staff alignment teams and can absorb compliance costs. This is the regulatory moat hypothesis, and it is the same logic that governs Layer 2 economics: dozens of rollups now compete for the same finite pool of users, and each justifies its existence with a security or scaling claim that, on inspection, slices liquidity rather than growing it. Fragility is the price of unsecured innovation, and the moat built from safety language is fragile precisely because it depends on the public believing the risk is real.
Which brings us to the trap in the other direction. If the safety narrative can be self-serving, so can its dismissal. Labeling warnings alarmist is a delegitimization move with a long pedigree โ tobacco, fossil fuels, and now compute. It clears the runway for expansion by making the precautionary position seem hysterical rather than evidence-based. The honest reading is bidirectional manipulation: both the alarmists and the dismissers have positions to defend, and neither has offered this market a falsifiable claim. Liquidity is a ghost, but the debt is real โ and the debt here is the trust deficit that accumulates when verification is deferred.
The contrarian conclusion, and the one I would defend against my own earlier skepticism, is that AI deregulation and crypto verification are not opposing forces but complements. The decoupling thesis most analysts carry โ that a light-touch AI regime starves the safety industry โ misses where demand actually originates. Compliance demand is regulatory. Verification demand is commercial. A bank integrating a model into its underwriting does not verify because a rule tells it to; it verifies because an unverifiable model is an unhedgeable liability. When the state steps back, the liability does not vanish. It migrates to the balance sheet of whoever signed the transaction. That migration is the market.
The caution is that verifiable compute is itself becoming a narrative commodity. In a bear market, every infrastructure thesis is eventually stress-tested by the same question: who is paying, and for what? Beyond the illusion, the current never truly stops โ capital keeps moving, but it moves toward cash flows, not toward frameworks. A verification layer with no paying counterparty is a whitepaper. A verification layer embedded in a settlement path is a business.
The macro frame ties this together. We are in a cycle where liquidity is scarce, where the cost of carrying unverified risk is rising, and where the institutions that once absorbed that risk are retreating. In that environment, the durable infrastructure is the infrastructure that replaces a withdrawn intermediary rather than merely critiquing it. Verifiable compute, on-chain attestation, and provenance registries are not exciting. They are not a narrative that pumps. They are the plumbing that lets two parties transact when neither trusts the other and no regulator will referee. That is the same function DeFi claimed, and the same function it largely failed to deliver at scale.
The signal to watch is not the next Treasury statement. It is the volume of verification demand that shows up in settlement flows, in insurance pricing, and in the compliance budgets of firms that cannot afford to be wrong. When the flow stops, we see what truly holds. In the quiet aftermath, only the resilient remain โ and resilience, in a deregulated market, is spelled v-e-r-i-f-i-e-d.