Last week, two of the most powerful artificial intelligence companies on earth did something this industry almost never does. They asked to be restrained.
OpenAI and Anthropic โ direct competitors, commercial rivals in nearly every respect โ sent UK lawmakers the same message: pass binding rules. Not principles. Not voluntary commitments. Not another "framework for responsible innovation" that everyone signs and no one enforces. Rules with enforcement behind them, the kind that carry penalties, audit obligations, and conditions on market access. When the graph spikes, the soul remains quiet. The crypto market read the headline โ a thin Crypto Briefing note, relayed secondhand, without a single named spokesperson or a line of proposed text โ shrugged, and went back to watching a range that has now held for weeks.
That shrug is the story. In a sideways market, everyone is waiting for direction, scanning charts for a signal. The signal arrived this month, and it was not on a chart. It was in a letter two AI labs handed to Westminster, and it is not really about AI at all. It is a template. And the template is already being applied to us.
To understand why that letter matters, you have to understand the room it walked into. Britain spent the years after Brexit trying to sell itself as the AI-friendly jurisdiction โ the place where the labs would come, open offices, and stay, because the rules were light and the talent was dense. The 2023 white paper said it plainly: a "pro-innovation approach," with sector regulators nudging rather than legislating. The bet was that restraint would attract capital. Meanwhile the European Union passed the AI Act, a machine built to classify models by capability and burden them by risk tier. The United States swung the other way, retreating from its earlier executive order toward something looser and more permissive. Britain found itself squeezed between a safety reputation it had earned and an investment pitch it wanted to keep.
Then two of the largest labs in the world โ one with the strongest commercial instinct, one with the loudest safety narrative โ walked in and pushed. Pass binding rules. OpenAI and Anthropic do not agree on much. That they agree on this is the signal. When direct competitors converge on a regulatory position, it usually means the fight has moved past whether rules should exist and into what kind of rules they will be. That is the most consequential phase of any regulatory cycle, and it is the phase almost nobody outside the drafting room bothers to watch.
I want to be careful here, because the source material is thin. What we have is a headline and a summary: no named spokespeople, no text of the proposed rules, no date, no indication of whether the appeal carried specifics or stopped at principle. On its own, that is almost nothing, and I would not build a thesis on it alone. But the value of a signal is not its density โ it is its direction. Two of the four or five entities on the planet capable of training frontier models just told a major democracy to build a gate. The question that should keep every builder in this industry awake is not whether the gate gets built. It is who holds the key, and whether we are standing inside or outside when it closes.
I have watched a version of this before. In 2020, during DeFi Summer, I sat on the other side of exactly this table, and the argument I lost then is the argument crypto is about to lose again.
Let me start with the mechanism, because the mechanism is the whole story. This is not a conspiracy. It is a strategy, and strategies are legible once you know what to look for.
OpenAI and Anthropic have already paid for the machinery that "binding rules" would require of everyone. Anthropic has its Responsible Scaling Policy, a tiered framework mapping capability to safeguards. OpenAI has its Preparedness Framework, with its own thresholds and evaluations. Both maintain safety teams, red-team processes, evaluation pipelines, and governance staff โ the expensive, unglamorous infrastructure of being careful. For them, the marginal cost of a new rule is a line item on a spreadsheet. For a twenty-person open-source lab in a rented office, the marginal cost of the same rule is survival.
When you raise the cost of entry, you do not raise the floor. You raise the wall. This is the oldest move in regulated industries, and it is not even hidden. Banks supported Basel. Pharmaceutical companies supported FDA trial standards, because they could afford the trials and their smaller competitors could not. Airlines supported safety mandates that grounded the undercapitalized. In every case the incumbents framed it as public protection, and in every case it also thinned the field. The two things are not in tension. They are the same thing wearing different clothes.
I learned this lesson in the least likely place: a DeFi liquidity mining program I refused to launch as designed. The proposal on the table was a classic โ a triple-digit APY, paid in a native token, engineered to pull TVL from every competing pool. The investors in the room wanted the number. They wanted the graph. What they did not want to hear was that the number was a subsidy wearing a yield's clothing. Liquidity mining APY is essentially a project paying to rent its own TVL figure โ pull the incentive and the real users evaporate. The metric you buy with emissions is not a community. It is a bill, and eventually it comes due.
The AI labs' compliance argument runs on the same engine. Safety capability, once built, becomes a moat that competitors must pay to cross. The graph of "responsible AI adoption" spikes. When the graph spikes, the soul remains quiet. What nobody asks is who is paying for the spike, and who can no longer afford to enter the room.
Notice the words the appeal travels in: "national security" and "consumer protection." That pairing is not accidental. It is precision political engineering, aimed at two different audiences at once. National security speaks to the hawks โ the legislators worried about frontier models turned against states, about cyberattacks and bioweapons and information warfare. Consumer protection speaks to the advocates โ the ones worried about deepfakes, fraud, and privacy. Put them together and you have covered the two dominant poles of the AI safety debate. You have made opposition expensive, because who wants to be the lawmaker who voted against national security and consumer protection on the same afternoon?
I spent two years inside rooms where language was chosen this carefully, and I learned to read it. In 2025, I served as a technical advisor to a coalition of protocol engineers lobbying for clearer regulatory frameworks ahead of the Bitcoin ETF approvals. My job was translation โ taking cryptographic concepts and turning them into briefs a regulator could read without a glossary. I watched how a sentence gets built to survive a hearing. "Censorship resistance" becomes "network resilience." "Permissionless" becomes "open access." The words are not lies. They are load-bearing. And when you see "national security" and "consumer protection" bolted together in a single appeal, you are looking at a sentence engineered to be impossible to vote against. That is not a criticism of the content. It is a note on the craft โ and craft like this is always in service of a strategy.
Here is the uncomfortable part. When the entities most capable of satisfying a rule are also the entities most active in writing it, the rule tends to get written in their shape. This is not corruption in the crude sense. It is something quieter and more durable, and the governance literature has a name for it: regulatory capture. Watch the threshold. The most likely shape of "binding rules" is a capability trigger โ a compute threshold, a model-size line, an evaluation standard that only frontier systems trip. Set the line high enough and you have created a rule that governs exactly the systems the incumbents already run, and exactly none of the systems their smaller competitors are trying to build. The threshold is where the moat is built. Set it at the frontier and you have drawn a circle that happens to contain OpenAI and Anthropic and happens to exclude everyone else โ not by intention on the face of the document, but by arithmetic.
I have written about this cost structure before, in a different arena. Layer 2 proving costs are absurd โ the economics of generating validity proofs are brutal, and unless gas returns to bull-market levels, operators of ZK rollups are quietly bleeding money just to keep the lights on. When a single structural cost is that heavy, it stops being a line item and becomes a filter. Only the well-capitalized survive it. Only the ones who can burn for years while the market waits. Compliance cost behaves identically. It is a proving cost for the right to exist, and the proof is not that your system is safe. The proof is that you could afford to say so.
And notice what the appeal does not include. There is no mention of independent third-party audits with teeth. No mention of model-weight disclosure. No mention of incident-reporting obligations. No mention of the labs submitting to the very rules they are asking the government to impose. The appeal is directional: pass rules, restrain the field. The restraint is pointed outward, at everyone, including โ in principle โ the ones asking. In practice, the ones asking have already built the capacity to absorb it. The rule that binds everyone binds the small hardest.
The appeal is framed as universal. Its effects are not. If Britain writes binding obligations around model release โ duties on publishers, liabilities on deployers โ the routes that suffer most are the open ones. Meta's Llama, Mistral's releases, the long tail of community models and fine-tunes. A closed API with a compliance team behind it can absorb an obligation. A weight file published to a public repository cannot absorb anything; it either exists or it does not. The rule that is neutral on paper is not neutral in practice. It lands hardest on the part of the ecosystem that has no legal department to catch it.
I have spent my whole career on the side of the ecosystem that has no legal department. At Gitcoin, in 2017, I helped build the quadratic funding mechanism for public goods โ a system designed to weight contributions by the number of people behind them rather than the size of their wallets. The whole point was to make small, numerous, distributed voices count against concentrated capital. I audited more than fifty prototype contracts by hand in those years, checking that the code enforced fairness rather than merely performing it. I spent nights debugging vote-weighting algorithms, driven by the belief that technology should empower communities and not just investors. I believed then, with something close to faith, that code could hold the line where institutions would not.
I still believe it. But I have also watched how easily the line gets moved before the code is even written. The AI appeal is being drafted in a language โ "frontier model," "compute threshold," "deployer responsibility" โ that will determine which parts of the ecosystem are visible to the law and which are not. A rule aimed at "frontier models" does not see the open-weight ecosystem at all, except as a risk to be managed. The threshold does not just decide who is regulated. It decides who is seen.
Everything I have described is happening to AI right now. It already happened to crypto, and it is happening again, and most of us are arguing about the wrong thing while it does. The pattern in crypto ran like this. First, a period of wild, permissive growth โ DeFi Summer, the ICO boom, the NFT explosion โ where the absence of rules let everything and everyone in. Then a period of pain โ collapses, fraud, the Terra/Luna unwind โ that handed regulators their mandate on a plate. Then a period of consolidation, in which the survivors discovered that rules they could afford were rules their competitors could not, and quietly stopped opposing the framework they had spent years denouncing.
I lived through every act of that play. I watched the ICO boom from inside Gitcoin, and I watched the DeFi Summer incentive wars from a boardroom where I was the only person arguing that a sustainable ecosystem needs real users, not rented TVL. I consulted on NFT royalty enforcement in 2021 and refused to sign off on an update that would have penalized secondary-market creators โ because I understood, before it was fashionable to say so, that decentralization which does not include economic justice for creators is just a new landlord wearing a decentralized costume. And in 2022, I watched Terra and Luna collapse and felt the floor of my own convictions move. That period taught me something I have carried ever since: the most dangerous illusion in this industry is the belief that structure can be replaced by conviction. Algorithmic stability failed because the structure underneath was hollow. Conviction did not save it. Nothing saves a hollow structure.
Now watch the same arc forming in AI, and watch crypto's response to it. While the labs write their own rules, crypto is still having the debate it has been having since 2013 โ whether regulation is good or bad. That is the wrong question, and it is wrong for the same reason it was wrong in DeFi. The question was never whether the incentive program exists. The question is who the incentives are designed to reward. In crypto, we learned โ slowly, painfully โ that the answer to "who does this rule serve" matters more than the answer to "should this rule exist." The AI labs have already learned it. They are acting on it. We are still arguing about the existence of the gate while they choose the lock.
There is a second-order effect worth naming, because it is where the money actually moves. "Regulatory certainty" is becoming a valuation input. A company that can shape the rules it will be judged by has its policy risk discounted; a company that cannot has that discount widened. Compliance capability stops being a cost center and becomes an intangible asset โ a reason to fund the incumbent and hesitate over the challenger. This is the same dynamic that reshaped crypto after 2022, when the exchanges that could afford legal infrastructure consolidated the market the ones who could not had to leave. The AI cycle will not be different. It will just be faster, because the stakes are larger and the political will is already there.
Here is the part that will make some of you uncomfortable, and I think it is the truest thing in this piece. The instinctive crypto response to a story like this is defiance. Two centralized AI labs are asking a government for binding rules; therefore regulation is a plot; therefore we oppose it. I understand that instinct. I have felt it. It is also, in this case, exactly the reaction the strategy is counting on.
Because if the crypto ecosystem's entire contribution to the AI governance debate is "no rules," then the rules get written without us. The people in the room writing thresholds โ the labs, the large incumbents, the regulators who already understand frontier models because those are the only models they have been briefed on โ will define "AI" in a way that leaves decentralized systems, open-weight models, and community infrastructure entirely outside the frame. Not banned. Worse than banned: unseen. You cannot defend a system the law does not know exists.
The contrarian read is that crypto's loudest anti-regulation voices are doing the work of the moat for free. Every time the industry refuses to engage with the substance of a rule โ thresholds, scope, definitions, exemptions โ it cedes the drafting table to the only parties willing to sit at it. And the parties willing to sit at it are the ones with something to gain from the shape of the pen. I have seen this exact abdication before. I have watched a royalty mechanism get written to penalize the very creators it claimed to protect, and I have watched the people it hurt stay silent because they did not want to "legitimize" the process. Their silence did not stop the mechanism. It just removed the only objection anyone would have heard. The blind spot, then, is not that regulation is coming. It is that we keep treating the arrival of regulation as the end of the argument, when it is the beginning of the only argument that has ever mattered: not whether the gate exists, but who it opens for.
I will go further, and this is where I expect to lose some of you. The uncomfortable symmetry is that crypto's own leaders are already running the AI labs' playbook. The largest centralized exchanges welcomed clear rules because they could afford them. Stablecoin issuers have lobbied for frameworks that entrench their position. The industry that spent a decade shouting "code is law" has quietly discovered that the entities who write the law get to decide what the code may do. The moat is not an AI invention. It is a human one, and we taught it to them. When a Bitcoin narrative gets rebranded onto an Ethereum stack and sold as a "Bitcoin Layer 2," that is the same move in miniature โ the repackaging of one thing as another to capture a market that does not yet know the difference. The AI labs are not doing anything we have not done. They are just doing it with better lawyers and a bigger budget.
Which leaves the pragmatic test. If the gate is coming โ and it is โ the only question that pays is this: what do we ask for at the drafting table? Not "no rules." That ship is in the harbor. We ask for thresholds defined by capability rather than by entity. We ask for open-weight carve-outs and tiered obligations. We ask for independent verification rather than self-attestation, for incident reporting with real teeth, for a seat for researchers and civil society and not just for the companies that can afford lobbyists. We ask, in other words, for the same things the AI labs will not ask for โ because they do not need them, and we do.
When the graph spikes, the soul remains quiet โ and the graph of "responsible AI governance" is spiking right now, in a letter most of this market did not read. The soul that should be loudest is ours, because the template being written in Westminster this month will be copied into every jurisdiction that regulates the technology we build. The gate is being designed. The lock is being chosen. The only open question is whether the crypto ecosystem shows up to argue about the shape of the key โ or spends another cycle arguing about whether gates should exist, and discovers, too late, that it was standing outside the whole time.
