Security, in the language of capital markets, is not a feature. It is a price.
On a gray January morning, two United States senators — Maria Cantwell of Washington and Mike Rounds of South Dakota — sent a letter to the Trump administration urging a coordinated federal response to artificial intelligence cybersecurity. On its surface, the document belongs to the world of appropriations hearings and agency jurisdiction. Read it again, though, and the subject is not safety at all. The subject is the cost of capital. When a government decides which technologies are permitted to touch critical infrastructure, it is not writing a rulebook. It is repricing an entire asset class.
I have spent nine years watching that repricing happen in slow motion, first in monetary policy transmission, then in decentralized finance, and now at the seam where algorithmic intelligence meets algorithmic money. And I have learned that liquidity is a mood, not a metric — it moves on the perception of the future long before the future arrives. Empathy for the humans on the other side of that repricing is not sentimentality; it is data.
Context: where the tide sits.
To understand why a Senate letter matters to markets that have nothing to do with the Senate, you have to map the present liquidity tide. Through 2025 and into 2026, the marginal dollar of institutional capital has been chasing one narrative: artificial intelligence as a general-purpose infrastructure build. Data centers, inference compute, power contracts, and the models layered on top of them absorbed capital at a pace that dwarfed every prior technology cycle. Crypto did not sit outside that tide. It floated on it, through AI-adjacent tokens, through decentralized compute markets, and through the quiet integration of automated strategies into derivatives venues.
The parallel that matters here is not the technology. It is the procurement standard. A decade ago, cloud providers discovered that the United States federal government — one of the largest buyers of computing on earth — would not purchase services that failed a specific security benchmark. That benchmark, FedRAMP, became a de facto global filter. Vendors that passed it gained a moat; vendors that ignored it were quietly excluded from the most stable revenue pool in the industry. What Cantwell and Rounds are proposing is FedRAMP for intelligence itself.
And the signal quality is high, because the pressure is bipartisan. In a capital cycle, a policy position held by one party is a trade. A policy position held by both parties is a regime. Structure is the skeleton; liquidity is the blood, and regimes are what determine where the blood flows for the next decade.
Core: what a certification regime actually does.
Here is where most commentary goes wrong. It treats AI security standards as a brake. In practice, mandatory certification behaves like a filter — and filters do not reduce volume. They reallocate it.
Let me be concrete, because this is where my own work becomes relevant. In March 2024, I sat with three senior portfolio managers at a Warsaw asset management firm to model something narrower than AI policy: the potential inflow of fifteen billion dollars of institutional capital into spot Bitcoin following ETF approval. We built liquidity-shock scenarios and stress-tested them against on-chain velocity. The exercise taught me a durable lesson — when a compliance gate opens, capital does not trickle through it. It floods, and it floods toward whatever already satisfies the gate. The gate itself becomes the moat.
The AI security push will do the same thing, but the gate is narrower. If safety certification becomes a precondition for federal procurement — and the bipartisan framing suggests it will — then the demand curve for compliant models turns inelastic. Buyers stop optimizing for raw capability and start optimizing for admissibility. In my own modeling of that shift, I estimated a five to ten percent incremental demand for inference compute, because safety evaluation and red-team testing are inference-heavy activities. That is not a rounding error. It is a new line item in the compute supply chain.
There is a measurement problem buried in all of this. When my colleagues and I modeled institutional ETF inflows back in 2024, the traditional macro frameworks we inherited — flows, positioning, funding rates — consistently understated what was happening because they could not see on-chain velocity: the speed at which tokens move between custodians, venues, and yield contracts. AI-driven capital inherits that blind spot and magnifies it. An algorithm does not wait for a monthly report to reprice risk. It reads the order book and the mempool simultaneously. If safety standards are calibrated against traditional risk models, they will regulate a market that no longer exists by the time the ink dries.
There is a second-order effect that almost nobody is pricing. In 2025, while auditing the regulatory compliance frameworks of five major staking providers ahead of Europe's MiCA implementation, I watched roughly five hundred million dollars in staked assets get reclassified as securities. Nothing about the underlying technology changed. The assets did not move. Only the legal category moved — and the risk profile of every holder moved with it. Compliance is not a constraint layered on top of an asset. It is a property of the asset.
Apply that lens to AI and the picture sharpens. Certification will not merely decide which models can be sold to the government. It will decide which models can be sold, full stop, because federal standards have a way of becoming the global default. This is where the fragmentation risk lives. The United States is building one track. Europe's AI Act is building a second. China's algorithm registry is building a third. For a multinational AI or crypto-AI firm, that is not one compliance cost. It is three, and they do not overlap. Illusions fade when the tide of liquidity recedes — and jurisdictional fragmentation is precisely the kind of structural friction that reveals itself only when capital gets scarce.
I know this pattern intimately. I have spent years arguing that the proliferation of Layer2 networks is not scaling — it is slicing an already-thin pool of users and liquidity into ever-smaller fragments. Compliance regimes are the Layer2 problem at the level of nation-states. Each one promises safety and scale; each one, in practice, splits the same finite capital across more walls.
Contrarian: the decoupling nobody expects.
Now the part that runs against the room. The reflexive read on regulatory pressure is bearish — more rules, less innovation, capital retreats. I think that read is backwards, and the blind spot is instructive.
Historically, when a compliance gate opens for a genuinely new asset class, the market misprices the event as a ceiling when it is actually a floor. The spot ETF approval was widely framed as "sell the news." It was not. The gate did not cap the asset; it admitted a class of buyer that had previously been structurally unable to participate. Regulatory clarity is a liquidity magnet dressed as a constraint. Patterns repeat, but the context never does — and the context here is that AI safety standards will convert a sprawling, unvetted market into a certified one, which is exactly the precondition institutional allocators require before they scale exposure.
But the deeper contrarian point is darker. The real risk is not over-security. It is asymmetric enforcement. If liability for unsafe outputs attaches to whoever ships the model, then closed, controllable systems will satisfy certification far more easily than open ones. Open models will inherit the risk without capturing the premium. That is not a safety outcome. It is a competitive moat wearing a safety badge.
I reached a version of this conclusion in August 2026, when I published a white paper arguing that AI-driven trading algorithms had captured roughly sixty percent of high-frequency liquidity in crypto derivatives markets. The criticism I received was that I was being a techno-pessimist. The fairer criticism is that I underestimated how much safety regulation would entrench those algorithms rather than restrain them. Only well-capitalized desks can afford certification-grade compliance. The standard does not slow the machines. It selects which machines are allowed to run.
Takeaway.
So watch two signals, not the headlines. First, whether the American AI Safety Institute publishes draft standards that are testable and quantified, or merely aspirational — the difference determines whether certification becomes a real filter or a paperwork ritual. Second, whether compliance quietly becomes a precondition for AI firms to raise capital or list publicly. That is the moment the moat hardens.
We are in a bull market, and bull markets are generous with illusions. The future is written in the present liquidity. The question is not whether Washington will write a safety standard. It will. The question is who holds the pen — and whether the capital that waits behind that gate will find an open door or a toll booth.
