On a gray Wednesday in Stockholm, I read a headline that contained two words I have learned to distrust in equal measure: reportedly, and policy center.
Ron Conway — the investor Silicon Valley calls its godfather, a man whose portfolio touches hundreds of companies — has reportedly tapped Jay Carney to lead a new organization focused on artificial intelligence policy. Carney is not the obvious pick for an ethics institute. He was White House press secretary from 2011 to 2014, and then spent seven years running global corporate affairs at Amazon, which is where the public record of his corporate career actually points, whatever the secondary coverage says about Google. He is, in other words, a professional translator. He spent a decade moving between the dialect of the West Wing and the dialect of the shareholder meeting, and he learned both without an accent.
That is a rare and valuable skill. It is also, depending on who signs the paychecks, a structural conflict of interest wearing a very good suit.
The item carried no primary source. No filing, no mission statement, no funding disclosure, no governance charter. Two facts and a qualifier. In nine years of tracking how narratives get manufactured long before they surface on a chart, I have learned that this silence is usually the loudest part of the story — it is the point where you start tracing the ghost in the machine.
To understand why this matters, you need the shape of the vacuum it is filling.
The United States still has no comprehensive federal AI statute. The EU AI Act passed in 2024 and is phasing in obligations on a timetable, which means European compliance officers are already drafting model cards and risk classifications while American firms wait for clarity that has not arrived. In the absence of federal law, the work has scattered to the states: Colorado's AI Act, California's patchwork of disclosure rules, a growing pile of procurement standards and algorithmic auditing requirements. Above it all floats NIST's AI Risk Management Framework — voluntary, technically serious, and unenforceable.
Voluntary frameworks are where good intentions go when nobody has the authority to enforce them. I watched the identical pattern in crypto between 2018 and 2022: industry associations publishing self-regulatory principles while the actual rules were written through enforcement actions. The lesson was never that self-regulation is worthless. It was that self-regulation without forced disclosure is a press release with a logo.
Into this vacuum walks a venture-funded policy center. There is precedent for the model. SIFT shaped American telecom policy through the 1990s. CCIA has done the same for technology for decades. The Partnership on AI, the AI Now Institute, and the Center for AI Safety occupy the academically credible end of the spectrum, staffed by people whose reputations rest on published research rather than portfolio returns. Conway's project sits at the other end, and on the evidence available, it is not pretending otherwise.
Here is the mechanism I want to name. Not lobbying, exactly. Something subtler and harder to litigate: the translation layer.
AI policy has a language problem. Legislators cannot read a loss curve. Engineers do not speak in statutory definitions. The people who can operate credibly in both registers are few, expensive, and permanently employable only by organizations with capital. That is not corruption. It is arithmetic, and arithmetic is harder to prosecute than bribery.
I spent sixty hours in late 2017 reading the Solidity of a token sale called Ethos, line by line, and finding three re-entrancy vulnerabilities before their public launch. I was thirty-two, I had a cybersecurity degree and no institutional backing, and I published the breakdown for free. What I learned was not that the code was malicious. It was that the code was unaudited — and that nobody in the crowd had an incentive to look. The incentive structure surrounding a fundraising event points away from scrutiny by design.
Policy has the same architecture. A center funded by the industry it advises has no re-entrancy guard, because the guard would be a funding disclosure, a conflict-of-interest policy, and a governance board with actual authority to remove its own director. Absent those three things, you do not have independent research. You have a translation layer with a salary attached. Code is law, but trust is fragile — and trust does not survive an unstated cap table.
Then there is the question of what a dual résumé actually produces once it is seated at a drafting table. In 2020 I co-authored a report with three independent researchers on Compound's governance, titled "The Illusion of Decentralization." We found nothing malicious. We found admin keys — a small set of addresses holding upgrade authority over a protocol marketed as autonomous. The finding was not that Compound was a lie. It was that the word "decentralized" had been used to describe a system whose failure modes had never been enumerated. Nobody had asked the boring questions: who can change this, how fast, and what happens when they do?
Applied to an AI policy center, the boring questions are the only ones that matter. What is its legal form — a 501(c)(3), a trade association, an LLC? Who funds it, and do funders hold agenda-setting rights over the research agenda? Are donors published? Is there a recusal policy when a member company's model sits at the center of the regulatory question being studied? None of that appears in the reporting, because the reporting contains two facts and a qualifier.
This is where my own work comes back around. In 2026 I led a small team evaluating the convergence of AI and blockchain ecosystems — decentralized compute markets, model provenance, inference verification. The report we published argued something narrower than the prevailing hype. Blockchain's gift to AI is not decentralized compute. Compute is centralized for sound engineering reasons; the capital intensity demands it, and pretending otherwise is how you end up with a GPU marketplace that routes through three custodians. The gift is the audit trail. A ledger gives you a machine-verifiable record of which data trained a model, which version was deployed, what prompt produced what output, and who signed off. That is a provenance layer, and it is the only part of the AI stack that regulation can actually grip.
But the audit trail argument has an uncomfortable inverse that the industry does not like discussing. If AI systems can be made legible, they can also be made selectively legible. Disclosure requirements can be scoped to death. "Transparency" can be defined down to a model card that merely confirms the model exists. And the drafting of those definitions is precisely the work a policy center does — the work that determines whether the audit trail covers the model's weights or just its marketing page.
The scarce resource in the next regulatory cycle is not compute and not capital. It is authenticity: the credibility of anyone claiming to speak for the public interest while simultaneously holding a fiduciary duty to a portfolio. Crypto learned this the hard way, and it is worth stating without decoration. Compliance-first institutions are not the same thing as decentralized ones. USDC can freeze an address in under a day. That is a permission slip with a stablecoin attached. An industry-funded policy center is the governance equivalent — fast, well-connected, useful, and revocable by whoever writes the checks. Calling it independent would require evidence of independence, and the entire point of a regulatory vacuum is that such evidence remains optional.
There is one more layer that crypto readers keep missing, and it is the one that will cost them the most. AI regulation will define what an "autonomous agent" is. That definition will determine whether on-chain agents — contracts that hold value, negotiate terms, and execute without a human signature at the moment of action — are legal persons, software tools, or financial intermediaries. Whichever bucket the definition lands in will reshape DeFi more thoroughly than any rule written about tokens directly. And there is no on-chain constituency in the room where that sentence is being drafted.
Bear markets make all of this worse, not better. In a drawdown, lobbying is a fixed cost and narrative is the only free asset. When token prices fall seventy percent, as mine did in 2022, the entities that survive are the ones with runway, and the entities with runway are the ones who can afford a permanent seat at the translation layer. The market does not reward the loudest voice. It rewards whoever is still in the building when the drafting ends. Listening to the silence between the blocks means noticing which conversations have no participants at all.
Here is the uncomfortable counterargument, and I hold it sincerely rather than as a debating device.
The alternative to an industry-funded policy center is not neutral governance by disinterested scholars. It is a vacuum, and vacuums get filled by the least coordinated option available. In the American case, that option is fifty state legislatures writing incompatible definitions of "high-risk AI," plus procurement rules that lock in whichever compliance vendor arrived first. Fragmentation is a tax on small builders and a subsidy to incumbents who can afford fifty compliance teams. A centralized, well-connected, industry-aligned center might at least produce something coherent enough to argue with in public.
There is a second angle that is more uncomfortable for my own tribe. The real capture risk is not OpenAI or Anthropic buying a seat at the table. It is the investor layer — the LPs and funds whose exit horizon is three to five years, operating against a policy horizon of ten. A center structured to protect portfolio value on a fund cycle will systematically prefer regulatory ambiguity, because ambiguity is where returns live and clarity is where margins compress. That preference is much harder to disclose than a donor list, because it is never written down anywhere.
The question is not whether Jay Carney is qualified. He is, and the translation gap he fills is real. The question is whether anyone is keeping a ledger of what the translation layer says privately and argues publicly — because the audit trail of broken promises gets written in exactly that gap, and right now nobody is holding the pen.
If the center launches without a disclosed funding structure and a charter that can actually remove its own leadership, treat the word "independent" the way I learned to treat the word "decentralized" in 2020: as a claim requiring evidence rather than a description of reality. Ask who holds the admin keys. Nobody has published them yet, and the silence is doing the answering.