I remember watching the liquidity dry up in the 2022 bear market, not because of capital flight, but because of a crisis of trust. Projects with multi-sig wallets controlled by a single entity collapsed overnight. The market didn’t just lose money—it lost faith in governance. Now, Anthropic, the AI safety darling, is planning super-voting shares for its CEO ahead of a potential IPO this year. On the surface, it sounds like a betrayal of every decentralization principle I’ve spent years advocating. But let’s dig deeper.
We didn’t build a future; we built a mirror. The mirror reflects our deepest anxieties about power concentration. Anthropic’s move to grant Dario Amodei outsized voting power is a mirror reflecting the tension between rapid AI deployment and long-term safety. Is this a step toward authoritarian control, or a pragmatic architecture for institutional trust?
Context: The Protocol and the Paradox
Anthropic is structured as a public benefit corporation with a long-term benefit trust. The trust holds a majority of voting rights, legally binding the company to prioritize safety over shareholder profits. Super-voting shares for the CEO are not new—Google, Facebook, and Snap have them. But in the AI sector, where the product is essentially a black-box oracle, governance becomes existential. The core question: Can a centralized governance model—even one bound by a benefit trust—deliver the decentralized trust that blockchain advocates demand?
From my experience auditing DeFi protocols, I’ve seen how "code is law" fails when the code is too complex for anyone to audit. Uniswap V4’s hooks, for example, turn the DEX into programmable Lego, but the complexity spike scares off 90% of developers. Similarly, Anthropic’s super-voting shares create a governance hook that centralizes decision-making power in one person. The risk is obvious: a single point of failure in judgment. But the benefit is equally clear: a clear, accountable figurehead for regulatory and institutional relationships.
Core: Mining for Truth in the Noise of IPO Mania
Let’s break down the technical and sociological implications. In blockchain, we talk about "trustless" systems—mathematical consensus replaces human judgment. But an AI model is not a blockchain; it’s a probabilistic engine. Its outputs cannot be verified by a smart contract. Therefore, trust must be placed in the governance structure that controls the model’s training, deployment, and updates.
Anthropic’s super-voting shares are a form of institutional trust architecture. The CEO, bound by the benefit trust, becomes the single point of accountability. This is not unlike the role of a blockchain foundation’s executive director in a delegated proof-of-stake system. The difference is that in crypto, we can exit the system by forking or switching to a competing protocol. In AI, model switching is expensive and risky. The barrier to forking an AI model is orders of magnitude higher than forking a blockchain.
Based on my experience with the Gnosis Safe multisig wallet during the 2022 crash, I learned that robust infrastructure is boring. Security comes from simplicity, not complexity. Anthropic’s governance is simple: one person responsible. But that simplicity is a double-edged sword. It provides clarity for regulators and investors—a clear point of contact for compliance. It also provides speed of decision-making, crucial in an AI arms race. However, the cost is the loss of collective intelligence.
The contrarian angle here is that Anthropic’s super-voting shares might actually be a pragmatic compromise for the current regulatory environment. The EU AI Act requires clear accountability for high-risk AI systems. A single accountable CEO is easier to regulate than a distributed autonomous organization. From a blockchain perspective, we hate this. But from a systemic safety perspective, it might be the only way to get AI into the hands of institutions that demand a human to sue if something goes wrong.
Contrarian: The Pragmatic Test
Let me apply the pragmatic test I developed during my "Trust Layer" framework work with EU banks. The test asks: Does this governance structure enable the system to survive a catastrophic failure?
In a blockchain, a catastrophic failure (e.g., a protocol exploit) is handled by code—a fork, a revert, a pause. The community can coordinate patch deployment. In an AI system, a catastrophic failure (e.g., a model generating harmful outputs at scale) cannot be patched instantly. The model must be shut down, retrained, or replaced. That requires a single decision-maker with authority to pull the plug. A distributed governance model could delay that decision by hours or days, during which the harm propagates.
Anthropic’s super-voting shares, therefore, function as a emergency circuit breaker. The CEO has the power to halt operations without consulting a board. This is analogous to a "kill switch" in a smart contract, but with a human judgment factor. The risk is that the human makes the wrong call—or refuses to pull the plug due to financial incentives. The benefit trust mitigates that by legally binding the CEO to safety, but the legal binding is only as strong as the enforcement mechanism.
The real question is: Can we audit the CEO’s decisions? In blockchain, every transaction is auditable. In Anthropic’s governance, decisions are made behind closed doors. The super-voting shares create opacity. This is the blind spot. Institutional investors may demand transparency, but the structure inherently limits it. The market must decide whether the CEO’s track record and the benefit trust’s legal constraints are sufficient substitutes for on-chain verification.
Takeaway: A Vision Forward
Open source is not a license; it’s a state of mind. Anthropic’s governance is not open source; it’s a closed trust model. But perhaps the future of AI governance is not about choosing between centralized and decentralized—it’s about building hybrid systems that leverage the strengths of both. Imagine a structure where the CEO has super-voting rights for operational decisions, but critical safety changes require a decentralized vote of a community of domain experts. This would combine accountability with collective intelligence.
As we approach the Anthropic IPO, I’ll be watching not the valuation, but the governance amendments. Will they include any mechanism for community oversight? Will the super-voting shares be permanent or sunset after a safety threshold is met? The market is right to be cautious. But dismissing the entire structure as anti-decentralization misses the point. We are mining for truth in the noise of AI mania. The truth is that governance is an evolving art, not a binary choice.
Liquidity isn’t just capital; it’s the flow of trust. Anthropic is betting that a single, accountable leader can channel that trust more effectively than a diffuse committee. Whether that bet pays off depends on the leader’s judgment and the strength of the legal bonds. For blockchain natives, this feels like a step backward. For the institutions that will adopt AI, it might be the only step forward.
— Root: Trust architecture is not about removing humans; it’s about designing for human fallibility. Anthropic’s super-voting shares are a design choice that prioritizes speed and accountability over distribution and auditability. The question is whether the market will price that trade-off correctly. I suspect it will, but only after the first stress test. And when that test comes, the response time will matter more than the votes.