I read the Crypto Briefing item on a Tuesday night in Chengdu, three tabs deep into a governance dashboard that had stopped moving. Australian Prime Minister Anthony Albanese had urged the United States and China to cooperate on AI risks. That was the whole story. A call, no mechanism, no timeline, no definition of the risk it wanted managed, no mention of who would be permitted to define it.
My chest tightened in a way I have learned to trust. I have seen this sentence structure before. In 2020, during DeFi Summer, I led a governance working group at MakerDAO and watched whale delegates urge "the community" to cooperate on risk parameters. Same grammar. Same fog. The word cooperate did the work of a thousand unstated assumptions, and the people who paid for those assumptions were the ones who never got a vote.
This is not really a story about artificial intelligence. It is a story about who gets to author the rules of a system that everyone must live inside. Curating the soul in a world of derivative clones begins with naming that.
Context: The Moment Governance Went Multilateral
Albanese's appeal did not arrive in a vacuum. It landed inside a specific, cluttered calendar. In November 2023, the United Kingdom hosted the Bletchley Park AI Safety Summit, the first serious attempt to pull frontier AI governance out of bilateral back-channels and into a room with chairs for everyone. In May 2024, South Korea convened the Seoul AI Summit, extending that table. By early 2025, France was hosting its own AI Action Summit in Paris, trying to fold in the Global South and the open-source community that Bletchley had largely left outside the door.
Three summits in roughly eighteen months. That is the unmistakable shape of a governance regime moving from conversation to institution. And whenever a governance regime formalizes, a question emerges that is older than blockchain and older than the nation-state: who sits at the table, and who is the table built to serve?
Australia's domestic record matters here, because a country cannot credibly broker abroad what it has not attempted at home. The Department of Industry, Science and Resources released a Proposals Paper in 2023 for the responsible use of AI, then moved toward mandatory guardrails โ the legislative scaffolding that turns voluntary principles into enforceable obligations. Australia has, in other words, skin in the game. It is not a spectator waving from the balcony. It is a mid-sized actor with a small but real governance workshop of its own, which is precisely the profile that earns the right to call for a bigger table.
The diplomatic track adds another layer. Albanese's government came to power in 2022 and, over the following two years, walked China policy back from the confrontational posture of the Morrison era โ a state visit in 2023, the gradual unwinding of trade sanctions through 2024. That thaw was not merely commercial. It was a repositioning. Australia, a bedrock AUKUS partner and therefore a technology ally of the United States, simultaneously rebuilt an economic channel to Beijing. That dual position is not a contradiction to be resolved. It is an asset to be monetized, and AI governance is the new market where such assets trade.
So when a middle power with a domestic framework, a security alliance with one superpower, and a repaired trade relationship with the other stands up and asks those two superpowers to cooperate, it is not being naive. It is executing the only play a middle power has. It is positioning itself as the middleware of the next governance stack.
Core: What On-Chain Governance Already Taught Us About This
The Middle Power Is Just a Mid-Size Delegate
I have spent the last several years inside DAO governance, and the structural lesson is uncomfortably portable. In any decentralized system, the loudest voices belong to the largest holders โ the whales. The smallest holders are numerous but atomized, their votes diluted to statistical noise. And then there is the middle: delegates with enough stake to be invited into working groups, enough reputation to be quoted, but not enough weight to dictate outcomes alone. Those mid-size delegates almost never win by force. They win by becoming indispensable. They draft the proposals nobody else wants to write. They mediate between two whale blocs that refuse to speak directly. They become the connective tissue.
That is the role Australia is auditioning for in AI governance, and it is the most durable role available in a two-polar system. A middle power cannot out-compute the United States or out-manufacture China. It can, however, become the neutral venue, the honest broker, the place where drafts get written and red lines get tested before they harden into ideology.
I saw this dynamic up close when I helped design the governance structure for CivicChain, a DAO focused on municipal data sovereignty. The hardest work was never the smart contract. It was the six months of translating between government regulators who spoke in statute and developers who spoke in commits. Neither side trusted the other's vocabulary. My entire value, professionally, was being the interpreter โ the person who could hear "data minimization principle" and "zero-knowledge proof" as the same sentence wearing different clothes. Australia is attempting the identical translation at a planetary scale.
Standards Fragmentation Is the Multi-Chain Problem
Here is where the crypto analogy stops being decorative and becomes analytically load-bearing. Anyone who has ever shipped across multiple chains knows the true cost of fragmentation. It is not the gas. It is not the bridging. It is the compliance overhead of maintaining two mental models, two security assumptions, two sets of audits, two definitions of what counts as a final transaction. Writing for Ethereum and then porting to a Bitcoin-adjacent environment means re-deriving your threat model from scratch. The cleverest teams survive it. The rest bleed out quietly, one integration at a time.
AI developers are about to inherit that exact tax. Right now, a company building a frontier model must satisfy American export controls, Chinese data-security law, and a growing patchwork of European obligations โ simultaneously, in the same codebase, under penalty of law in each jurisdiction. The rules do not merely differ. They conflict. And when rules conflict, engineers do not choose the safest path. They choose the path with the least litigation exposure, which is often the least transparent one.
If the United States and China were to converge โ even narrowly โ on shared evaluation standards, red-teaming protocols, or provenance labeling for generated content, the fragmentation tax would fall for everyone operating in third markets. The beneficiaries of AI cooperation are not primarily the two superpowers. They are the thousands of developers who currently live in the crack between two incompatible rulebooks.
This is the part of the conversation that gets lost when the framing is reduced to great-power rivalry. The interesting question is not whether Washington and Beijing will become friends. They will not. The interesting question is whether the interoperability layer of AI governance gets built, and if so, by whom.
The Regulation That Nobody In The Room Wrote
I need to bring in something older and darker, because it is the reason this story unsettles me more than a policy analyst might expect.
In August 2022, the U.S. Treasury sanctioned Tornado Cash, a privacy protocol, and โ critically โ added wallet addresses associated with its immutable smart contracts to the Specially Designated Nationals list. I want to be precise about what that meant, because the imprecision in most coverage is where the danger hides. The sanction did not merely target a business or a set of operators. It reached toward the code itself, and by extension toward the people who had written, deployed, and even interacted with it.
I have audited governance systems and reviewed smart contracts for years, and I can tell you with confidence that no developer I know sat down and asked, "Should I write code that a future regulator will criminalize?" We all assumed the boundary lay between using a tool and building a tool. Tornado Cash suggested the boundary might not exist at all. Writing infrastructure could be construed as facilitating whatever some downstream user did with it.
Now transpose that logic into AI. Every model is a general-purpose tool. Every paper published on alignment, every open-weight release, every red-team methodology is dual-use by definition. If the precedent holds that publishing the artifact is equivalent to committing the act, then AI safety research becomes legally radioactive, and the researchers most likely to leave are the careful ones. The reckless ones have nothing to lose. This is the inversion that should terrify anyone who actually cares about safety.
So when Albanese calls for cooperation on AI risk, I hear an opening that cuts two ways. If cooperation produces a shared framework that protects builders โ that draws a defensible line between authoring a tool and weaponizing it โ it is the most valuable diplomatic product available. If cooperation instead produces a shared enforcement apparatus without a shared theory of liability, it hardens the Tornado Cash logic into international norm, and the next decade of AI development happens in jurisdictions that promise the least accountability.
The Bear Market Is Where This Gets Personal
We are in a bear market, and I want to be honest about why that matters to a story that reads like geopolitics. In a bear market, the reader is not asking abstract questions. The reader is asking whether their assets survive.
Governance fragmentation is not an abstract risk to that question. It is the mechanism by which protocols die. Consider what happens to a decentralized protocol when two major jurisdictions disagree on whether a given activity is legal. The protocol cannot comply with both. So it either forks its functionality by geography, which fragments liquidity, or it retreats to the permissive jurisdiction and abandons the rest, which fragments its user base. Either way, liquidity thins. When liquidity thins, incentive programs lose their grip, and when incentive programs lose their grip, the protocol that was solvent last quarter becomes a footnote this quarter.
Over the past several cycles, I have watched mid-cap protocols lose more than a third of their liquidity not because of a hack or a bad trade, but because a regulatory boundary shifted under their feet and they had no interoperability layer to stand on. The loss was never announced. It just showed up, week by week, in the dashboard I keep open. The most dangerous force in a bear market is not volatility. It is incoherence. Volatility is survivable. Incoherence is not, because you cannot model a risk whose rules keep changing in different directions at once.
That is why I read Albanese's call not as a diplomatic courtesy but as a signal about infrastructure. Any movement toward interoperable standards reduces incoherence. Any movement away increases it. And the crypto reader, whether they care about AI or not, is ultimately exposed to which way this goes.
Contrarian: Cooperation That Consolidates the Blocs
Now for the part that the optimistic framing prefers to skip.
There is a version of "US-China AI cooperation" that is genuinely good โ and there is a version that quietly entrenches exactly the fragmentation it claims to solve. I have watched the second version operate inside crypto governance for years, and the pattern is reliable enough to be predictive.
Start with what actually happens when two large powers agree to "cooperate" on a sensitive technology. They almost never share the sensitive parts. What they share is the perimeter โ the definition of what is out of bounds, the list of who is excluded, the standards that everyone else must meet while the two of them continue competing everywhere it matters. Cooperation at the frontier becomes a gentlemen's agreement to freeze the frontier, while the actual races continue underneath.
This is the pattern I have come to call minimal non-hostility: a narrow truce on existential red lines that leaves the rest of the competition entirely intact โ and often more orderly, which means more durable. A truce that turns a chaotic arms race into a managed duopoly is not peace. It is a better-governed conflict. And better-governed conflicts are harder to exit.
I have seen this exact move in token governance. Two large blocs agree to "prioritize stability," and stability becomes the word under which the smaller holders are quietly excluded. In 2020, I published a dissenting analysis at MakerDAO arguing that the risk parameters favored large collateral holders while imposing the sharpest costs on the smallest โ and the counterargument I received most often was not that I was wrong, but that raising the issue threatened stability. The same word. The same function. A virtue deployed to protect a structure.
Here is the honest Blindspot. If US-China AI cooperation materializes, it will most likely look like the Trump-era and Biden-era export-control layers surviving intact โ the chip restrictions of 2022, 2023, and the tightening rounds that followed โ with a thin safety dialogue layered on top. The safety dialogue will be real, and it will be narrow. It will cover catastrophic risk, perhaps autonomous weapons, perhaps model-evaluation sharing. It will not cover market access, compute allocation, or the terms on which third countries participate. Those decisions will continue to be made by the two parties that show up. Everybody else will be consulted.
And now the second contrarian point, the one that makes this personal to a crypto audience. I have argued for years that nearly all of what markets call "Bitcoin Layer 2s" are Ethereum projects wearing a Bitcoin costume โ rebranding exercises that borrow the legitimacy of the older chain without inheriting its security model or community. The mechanism is identical to what is about to happen in AI governance. A framework built by, about, and for the two superpowers will be exported as "global AI standards," and everyone will be expected to adopt it because it has the word global on the label. The branding will do the work that the substance never earned. Curating the soul in a world of derivative clones means noticing when legitimacy has been borrowed rather than built โ and refusing to applaud the costume.
The falsifiable test is straightforward, and I want to state it plainly so that we can be held to it later. Watch for a specific mechanism: a standing working group, a shared evaluation repository, a joint red-team protocol, or โ most ambitiously โ an intergovernmental body modeled on the climate panel. Watch whether membership is genuinely multilateral or cosmetically so. Watch whether the outputs are binding or advisory. If, eighteen months from now, the only artifact of this call is a communiquรฉ, then the call was never about governance. It was diplomacy-as-image, and the middle power was middleware for a message rather than for a mechanism.
There is one more uncomfortable truth I owe the reader. The most plausible reason Albanese made this call is not pure idealism. It is hedging. Australia sits on the fault line between a security guarantee it cannot abandon and a trade relationship it cannot afford to lose. Full US-China confrontation would force a choice Australia desperately does not want to make. So the rational move is to advocate for a coordination layer that prevents the choice from ever becoming mandatory. That is smart statecraft. It is also a reminder that when a middle power speaks of the global interest, it is usually describing its own survival in respectable terms. The same is true of every delegate who ever wrote a "neutral" governance proposal.
Takeaway: The Table Being Built in Advance
I keep circling back to something I wrote during the 2022 bear market, when the numbers on my own screen had turned a color I did not want to name. I argued then that decentralization was a form of emotional security โ not a guarantee against loss, but a structure that lets you acknowledge loss without being destroyed by it. I still believe that. But the past three years have added a harder edge. A structure that cannot be governed coherently is not secure. It is merely unmanaged, and unmanaged things do not fall โ they get taken.
That is the real stake in Albanese's call. AI governance is being built right now, in real time, by parties who show up. The standards being drafted this year will determine the compliance surface of the next decade โ for AI companies, and, because the two fields are converging, for the on-chain economy that increasingly runs on inference, agentic systems, and model-gated applications. A developer launching anything in 2028 will inherit the governance stack being assembled today. Curating the soul in a world of derivative clones is no longer a metaphor about NFTs and provenance. It is the fight over whether the coming rules will have a soul at all, or whether they will be pure perimeter.
So I am not asking whether Australia should have spoken. I am asking what we do with the grammar it used. When a powerful actor urges "cooperation," the responsible move is not to cheer and not to sneer, but to ask the three questions we should have learned to ask in crypto a decade ago. Who wrote the first draft? Who is implied by cooperation, and who is quietly assumed out of it? And what is the falsifiable test that tells us whether this was governance or theater?
The table is being set. The middle power wants a seat at it. The two giants will decide how many seats exist. And the rest of us โ the builders, the small holders, the people who will live inside whatever this becomes โ will inherit the menu.
We were not invited to write it. We should at least insist on reading it before we are seated.