Anthropic's Super-Voting Shares: The Same Governance Trap, Different Industry

The logic held; the incentives were broken. Anthropic, the AI safety lab valued at over $30 billion, plans to grant super-voting shares to its CEO and co-founders ahead of its IPO. The announcement landed with the clinical precision of a smart contract deployment—clean, deterministic, and devoid of nuance. I traced the narrative to the governance mechanics. The structure is not new. It is the same dual-class arrangement that has propped up founder control at Google, Meta, and Snap. But in the context of an AI company that markets itself as a steward of responsible development, the implications are different. The yield was not profit; it was liquidity of power. Anthropic's pitch is simple: control stays with the mission-driven founders, insulating the company from short-term market pressure. The co-founders, Dario and Daniela Amodei, along with a handful of others, will hold shares with voting power ten to twenty times that of ordinary stockholders. The IPO, expected within the next eighteen months, will offer public investors a stake in the company's financial upside but not in its strategic direction. This is a common pattern. In 2017, I audited the smart contracts of three ICO projects that promised decentralized governance. Each had a multi-sig wallet controlled by the founders. The code did not lie, but it could be misled. The same principle applies here. The super-voting shares are the digital equivalent of a backdoor admin key—except this time, the contract is written in legal language, not Solidity. I have seen this before. In 2020, I spent four hundred hours tracing the incentive flows of Compound Finance. The governance token was a marketing tool, not a control mechanism. The real power sat with the team and a handful of venture capitalists. The yield was subsidized by inflation, not organic revenue. The logic held; the incentives were broken. Anthropic's super-voting shares are a more transparent version of the same illusion. The company will present itself as a decentralized stakeholder in AI safety, but the decision-making will remain concentrated in a wallet that holds a few private keys. The supply was fixed; the demand was fabricated. Let me dissect the mechanics. Super-voting shares are a class of stock that grants disproportionate voting power per share. Typically, founders receive shares with ten votes each, while public investors receive one vote per share. The structure is irreversible unless a sunset clause is triggered—a provision that automatically converts super-voting shares to ordinary shares after a set period or upon the founder's resignation. Anthropic has not disclosed its sunset terms. Based on my experience auditing DAO governance structures, the absence of a hard sunset is a red flag. It signals that the founders intend to hold power indefinitely. Code does not lie, but it can be misled. The absence of a sunset is a silent default. The implications for valuation are measurable. Institutional investors, particularly index funds and ESG mandates, apply a governance discount to dual-class companies. Research from ISS and Glass Lewis shows an average discount of 5% to 10% on the IPO price. For a company with a $30 billion valuation, that is $1.5 to $3 billion of value left on the table. But the discount is not the only cost. The super-voting structure also introduces a systemic risk: the concentration of decision-making in a small group of individuals who may be wrong about the direction of AI safety. The 2022 Terra/Luna collapse was a mathematical inevitability, not a black swan. I modeled the feedback loop three days before the crash. The algorithm was not flawed; the incentives were. The same principle applies to Anthropic. The founders are building a moat around their control, but they are also building a prison for accountability. Now, the contrarian angle. The bulls will argue that super-voting shares protect Anthropic's mission from the short-termism of public markets. They will point to Google's ability to invest in long-term projects like Waymo because of its dual-class structure. They will say that AI safety requires a concentrated vision, not a fragmented board. There is some truth to this. Algorithmic fairness assumes fair inputs. If the founders are genuinely committed to safety, the structure enables them to execute without interference. But the assumption is fragile. The yield was not profit; it was liquidity. The real question is not whether the founders have good intentions, but whether the governance structure allows for error correction. In my 2021 analysis of the Bored Ape Yacht Club mint, I discovered that the bot scripts were not hiding. They were using the same gas bidding patterns as the insiders. The system was designed to favor the connected. Transparency is a feature, not a default state. Anthropic's super-voting shares are transparent, but they are not accountable. I have been investigating blockchain governance for a decade. The pattern is always the same. The founders promise decentralization, then build a structure that centralizes control. The difference is the industry. In crypto, the governance tokens are often worthless. In AI, the governance tokens are shares of a company that could define the future of intelligence. The stakes are higher, but the mechanics are identical. The logic held; the incentives were broken. Anthropic's super-voting shares are a governance trap dressed in mission-driven language. The takeaway is not a conclusion but a call for accountability. The market will price this structure, but the price will not capture the second-order effects. The real risk is that the concentration of power will lead to a single point of failure in the AI safety ecosystem. I have seen this before. The 2017 ICOs collapsed because the team held the keys. The 2020 DeFi protocols drained because the admin key was not multisig. The 2022 Luna ecosystem vaporized because the algorithm assumed infinite growth. Anthropic is not a crypto project, but the governance trap is the same. The question is not whether the founders are good, but whether the system can survive their mistakes. The answer is in the code. And the code is not written yet.