The proof is silent; the code screams the truth. On August 19, The Information reported that Anthropic is preparing a class of stock with enhanced voting rights for CEO Dario Amodei and other co-founders. This is not a governance innovation. It is a structural admission: the founding team holds a low percentage of shares, and they fear dilution of control. The proposed mechanism—super voting shares—will allow the founders to elect a majority of board members, while retaining a non-shareholder trustee system. This is the same pattern I have seen in dozens of blockchain protocols: a privileged class of validators that can override the consensus of the majority.
I do not trust the contract; I audit the logic. Anthropic’s logic is explicit: protect the founding vision from external shareholders. But at what cost? The same logic that centralizes governance in a for-profit AI company mirrors the validator centralization I have analyzed in proof-of-stake networks. When Lido’s node operator set became too concentrated, the network’s security depended on trust in a few operators. Anthropic’s super voting rights are a concentrated node operator set for corporate governance. The difference is that in crypto, we can fork the protocol. In a private company, you cannot fork the board.
Context: The Protocol Mechanics of Anthropic’s Governance
Anthropic is structured as a Delaware public benefit corporation. Its governance model currently includes a non-shareholder trustee system, which is designed to balance profit motives with the company’s stated mission of safe AI development. The new super voting rights add a second layer of control: a special class of shares that gives the founders an outsized voice in board elections. This is analogous to a blockchain’s on-chain governance system where a small set of core developers control the upgrade mechanism. The difference is that blockchain upgrades are transparent and verifiable; Anthropic’s governance changes are opaque and contractual.
If you examine the capital structure, Anthropic has raised over $7 billion from investors like Google, Spark Capital, and Salesforce. The co-founders’ equity stake is likely below 10% after multiple funding rounds. The super voting rights are a defensive mechanism against shareholder activism. But from a cryptographic perspective, this is a failure of distributed trust. The founding team is essentially creating a multisig where they hold the vast majority of keys. The trustee system remains, but its influence is now subordinated to the super voting shares.
Core: Code-Level Analysis and Trade-offs
Let me break this down in terms of smart contract architecture. In a decentralized protocol, governance is often implemented via a token-weighted voting system. The most secure implementations use timelocks, quorum thresholds, and withdrawal delays. Anthropic’s governance is a permissioned blockchain with a centralized sequencer. The super voting rights are a hardcoded privilege that cannot be overridden by any external actor. This is a design choice that prioritizes the founding team’s vision over the principle of decentralized decision-making.
From my experience auditing DeFi contracts, I have seen similar patterns. In 2020, I analyzed the Compound Finance governance model and identified a vulnerability: the admin key could unilaterally change the interest rate model. That key was controlled by a multisig of the founding team. The theoretical risk of a malicious upgrade was low, but the structural risk was real. Anthropic’s super voting rights are the same as that admin key, but with a more permanent lock. The trade-off is clear: faster decision-making vs. censorship resistance. For an AI company, fast decision-making might be critical for safety. But the same mechanism can be used to suppress dissenting voices within the organization.
I want to quantify the risk. Let’s assume the founding team owns 8% of the equity. With super voting shares, they can control 51% of the board. This is a 6.4x leverage on their voting power. In crypto terms, this is equivalent to a single whale holding 51% of the staking supply. The probability of a governance attack increases exponentially with such concentration. The trustee system was supposed to be the check, but the special class shares effectively neuter it. The trustees are now like a minority validator committee that can never produce a block.
Contrarian: The Blind Spots of Trust and Verifiability
The counter-intuitive angle is that Anthropic’s governance structure might actually be more resilient than a fully decentralized model. Hear me out. In a startup, the founders have the deepest knowledge of the technology. If external shareholders push for short-term profit, they could force the company to release unsafe AI models. The super voting rights protect the long-term mission. But the problem is that this protection is opaque. There is no on-chain verification. You have to trust that the founders will act in good faith. The same trust that blockchain is supposed to eliminate.
This is the blind spot. The crypto community often celebrates governance decentralization as an absolute good. But my analysis of the 2022 bear market showed that decentralized governance can be slow and chaotic. During the Terra collapse, the Luna governance failed to act because of infighting. Anthropic’s model is the opposite: centralized control allows rapid response. The question is whether the rapid response is always in the interest of the network.
I recall my experience with the Zcash side-channel vulnerability in 2017. The core team had to make a quick decision to patch the proving system. They centralized the decision-making process, and it worked. But if the same team had been subject to a shareholder vote, the patch might have been delayed by weeks. There is a lesson here: for technical systems, a degree of centralization is necessary for security. The challenge is to design the escape hatch so that the centralization can be revoked if the trusted party becomes malicious.
Anthropic’s super voting rights have no escape hatch. The special class shares are perpetual. There is no mechanism for the majority of shareholders to revoke the super voting rights. This is a permanent lock that creates a single point of failure. If the founding team ever becomes compromised—through coercion, bribery, or simple human error—the entire governance structure is compromised. In crypto, we have timelocks and multi-sig recoveries. Anthropic has nothing but a contract that says “trust us.”
Takeaway: The Vulnerability Forecast
I see a future where this governance structure becomes a vector for attack. As AI agents begin to interact with financial systems, the control of AI companies will become a national security issue. A super voting share is a prime target for state-sponsored actors. They do not need to compromise the AI model; they only need to compromise the few individuals who hold the super voting shares. This is a 51% attack on the human level.
Consensus is fragile. Math is eternal. Anthropic has chosen a governance model that is efficient but fragile. The bear market is a good time to ask: whose assets are safe? For now, the answer is only the founders’. Investors should demand a verifiable governance mechanism, not a promise.