Anthropic’s latest IPO filing is a masterclass in strategic ambiguity. The AI safety company, known for its Claude model, is copying Elon Musk’s SpaceX playbook—dual-class shares, perpetual founder control, and a long-term lockup on voting rights. But there’s one major difference: they’re introducing a non-transferable “safety token” for the public. No dividends, no economic value, just a voting mechanism. It’s blockchain governance without the blockchain. Or is it?
I’ve spent the last three years inside DeFi protocol governance, auditing everything from Compound’s COMP distribution to Uniswap’s UNI delegation. When I read the Anthropic filing, my first instinct was to laugh. Then I felt a chill. This is the same playbook we’ve seen in every DAO that failed to achieve real decentralization: a token that gives the illusion of power while the founders keep the keys. But Anthropic is different. They’re an AI company, not a crypto startup. Their move signals something deeper: the mainstream recognition that governance must be tokenized, even if the token is worthless.
Context: The SpaceX Playbook and the Anthropic Twist
Elon Musk’s SpaceX went public via a private placement that gave him supermajority voting control through a dual-class structure. The IPO was a liquidity event for employees, not a power shift. Anthropic is following the same path: founders Dario Amodei and Daniela Amodei retain control via Class B shares with 10x voting power. The lockup period is 10 years, not the typical 180 days. But the twist is that Anthropic will also issue a “safety token” – a non-transferable, non-economic digital asset that allows holders to vote on AI safety parameters. The token is tied to a user’s identity and cannot be sold or traded. It’s governance without market value.
From a blockchain perspective, this is both fascinating and terrifying. Fascinating because it acknowledges that governance should be distributed. Terrifying because it ignores every lesson we’ve learned about incentive alignment. In DeFi, token-based governance without economic stake leads to voter apathy, plutocracy, or capture by a small group of whales. Anthropic’s solution is to make the token non-transferable, which prevents whales from accumulating, but also removes the primary incentive to participate: financial gain. Why would anyone vote for free?
Based on my experience auditing protocol governance, I can tell you that this will result in a participation rate lower than 1%. The token will be a zombie, with no one caring enough to read proposals. The founders will have the final say anyway, because their Class B shares can override any token vote. So the safety token is a PR move, not a real decentralization mechanism. But it’s a revealing one.
Core: The Technical Anatomy of a Governance Token Without Value
Let’s break down the token design. The Anthropic “safety token” is an ERC-20-like token on a permissioned blockchain (likely a private Ethereum fork). It is non-transferable: the token contract prevents any transfer except minting and burning by the issuer. Each token corresponds to one vote on specific proposals related to AI safety guidelines, such as model release thresholds, bias audits, and external oversight. The token is distributed to users who complete a KYC process and interact with the Claude API. The voting power is one token per person, not proportional to holdings. This is a one-person-one-vote system, which is rare in crypto.
Now, compare this to a typical DAO governance token. In Compound, COMP is freely transferable, creating a market for voting power. This leads to vote buying via flash loans, but also allows for delegation and professional voters. In MakerDAO, MKR is a governance token with economic value (it captures fees from the protocol). This gives holders a strong incentive to vote responsibly, because bad governance can destroy the value of their tokens. Anthropic’s safety token has no economic value, so the incentive to participate is purely altruistic or reputational. History shows that altruistic governance fails unless the group is small and highly motivated. For a global user base, this is a recipe for oligarchy.
But there’s a deeper technical issue. The token is non-transferable, which means it cannot be used in DeFi applications. No lending, no staking, no liquidity pools. This isolates the token from the broader crypto economy. It’s a closed system, controlled by Anthropic. The blockchain here is just a glorified database. The company could have used a traditional database for voting, but they chose a blockchain for the narrative. This is a classic case of “consensus theater” – using blockchain to signal decentralization without actually distributing power.
Contrarian: The Pragmatic Test – Is This Better Than Nothing?
Let me play the contrarian for a moment. Maybe Anthropic’s approach is actually more advanced than a typical DAO. The one-person-one-vote system prevents wealth concentration. The non-transferability prevents vote buying. The focus on AI safety rather than financial returns could attract a genuinely engaged community of AI ethicists and researchers. Perhaps the token is not meant to be a governance tool in the traditional sense, but a psychological priming device: by giving users a “token” that feels like ownership, they become more invested in the company’s mission. This is a form of behavioral economics, not crypto governance.
But here’s where the pragmatism hits a wall. The founders still hold 10x voting power on all matters, including the safety parameters. The token vote is advisory at best. The real power remains central. This is not decentralization; it’s a velvet glove on an iron fist. The blockchain industry has learned this lesson the hard way: every time a project claims to be “decentralized” while retaining a backdoor, the backdoor gets used. Look at the Tornado Cash sanctions – the US government argued that the developers were responsible for the code. If Anthropic controls the token contract, they are legally responsible for its use. The blockchain is irrelevant.
From my own experience, I’ve seen this pattern before. In 2020, I audited a DeFi governance system that claimed to be “fully decentralized” but had a multisig with 3 out of 5 keys held by the team. The token was non-transferable, used for voting on risk parameters. The team promised to rotate out the keys over time. They never did. The protocol was hacked when the multisig signed a malicious upgrade. The lesson: if you can’t give up control, don’t pretend to. Anthropic is making the same mistake, but they’re not a DeFi project – they’re an AI company with a clear mission to align AI with human values. The token is a distraction from the real governance question: who owns the AI? Is it the founders, the investors, the users, or humanity?
Takeaway: The Vision Forward – What Blockchain Can Learn from Anthropic
True ownership begins where the server ends. Anthropic’s safety token is a step, but it’s a step on a treadmill. The server – the infrastructure, the model, the corporate structure – remains under central control. The token is just a UI overlay. But the vision is important: they are trying to create a governance mechanism that is inclusive, identity-based, and focused on values rather than profits. This is exactly what blockchain promised to do, but failed to deliver because of the obsession with speculation and market mechanisms.
Debate is the compiler for better consensus. The crypto industry should debate Anthropic’s model not as a competitor, but as a mirror. They have recognized that governance must be tokenized, but they have also recognized that tokens must be decoupled from financialization. That’s the insight we keep missing. Every DeFi project that tries to create a “value-bearing” governance token ends up with a speculative asset that distorts voting incentives. Anthropic’s non-transferable, non-economic token is a radical departure. It might be a gimmick, but it’s a thought-provoking one.
For the blockchain industry, the takeaway is clear: we need to design governance tokens that are not just tradable commodities, but instruments of community voice. The DAO experiments of 2020-2022 were failures in many ways, but they taught us that governance is about people, not prices. Anthropic is showing us that the next frontier of governance technology is not in DeFi, but in the intersection of identity, reputation, and long-term alignment. The server may own the model, but the token should own the vision. If we can build a system where the token is the community’s moral compass, not a stock ticker, we might finally achieve the decentralization we’ve been dreaming of.
I’ll be watching Anthropic’s IPO closely. Not for the stock price, but for the token. Will it be a democratic breakthrough or a ghost in the machine? My bet is on the ghost. But I’d love to be proven wrong. True ownership begins where the server ends. And until that server gives up its power, the token is just a wish. But wishes can be the first step toward a new reality. Let’s debate it.