Centralized governance is the new liquidity trap.
Anthropic, the AI safety company, filed plans to issue super-voting shares to CEO Dario Amodei ahead of a potential IPO later this year. The structure gives Amodei 10x voting power per share compared to public investors. The news broke last week, sending a predictable ripple through the AI investment sector. Market commentators framed it as a standard founder-protection mechanism.
I see it differently.
Over the past 14 years, I have audited over 200 smart contracts, traced stolen funds from the 2xBT wallet breach, and manually reconciled FTX's on-chain liabilities. I have learned that governance structures are not just legal forms. They are attack surfaces. When a single entity holds disproportionate voting power, the system's security model collapses into a single point of failure.
Volatility is just liquidity leaving the room.
Anthropic's governance plan is not a departure from tech industry norms. Google, Facebook, and Snap all used multi-class stock structures to concentrate control in founders. But the context matters. Anthropic positions itself as a steward of safe AI development. Its founding narrative emphasizes alignment, transparency, and decentralized oversight. The super-voting share structure directly contradicts that narrative.
Context: The Hype Cycle Meets the Governance Reality
Anthropic was founded in 2021 by former OpenAI employees. Its mission: build AI systems that are interpretable, steerable, and trustworthy. The company raised over $7 billion from investors including Google, Salesforce, and Spark Capital. Its valuation crossed $18 billion in early 2024.
The IPO is expected to be one of the largest tech listings of the year. Investment banks are already positioning for fees. The AI sector is experiencing a liquidity glut, with VCs pouring capital into any startup with a language model.
Against this backdrop, Anthropic's super-voting share structure appears as a standard anti-takeover provision. Founders argue that it protects long-term vision from short-term shareholder pressure. In the crypto world, we call that the "centralization thesis."
Core: Systematic Teardown of the Super-Voting Model
Let me dissect the mechanics.
Anthropic plans to issue Class B shares with 10 votes per share, held exclusively by Amodei and a small group of insiders. Class A shares, sold to the public, will carry one vote each. The exact ratio may vary, but the principle is clear: control is not proportional to economic ownership.
Based on my audit experience with the Governor Bracelet contract, I learned that centralized control points are where risk concentrates. In that $12 million pool, a single reentrancy vulnerability existed because the developer had unfettered access to upgrade the contract without community vote. The fix was simple: remove the super-admin key. But the team refused, citing the need for "rapid response." The exploit happened three weeks later.
Anthropic's super-voting shares are the same pattern. The CEO gains the ability to override shareholder votes on critical decisions: board composition, executive compensation, mergers, and even the company's core safety commitments. The alignment between economic incentives and governance rights is broken.
Trust is a variable I refuse to define.
In the crypto space, we have seen this movie before. Uniswap V4's hooks introduce programmable complexity, but the governance layer remains a bottleneck. The key difference is that Uniswap's governance token holders can vote to upgrade or veto changes. In Anthropic's model, the CEO holds an unassailable majority. There is no on-chain recourse.
Let me quantify the risk.
Assume Anthropic raises $5 billion in the IPO at a $20 billion valuation. The public float will be approximately 25% of total shares. If Amodei holds 15% of the economic interest but all of his shares are Class B, his voting power would be approximately 60% of total votes. That is a veto-proof majority. He can pass any resolution, including a resolution to redact safety commitments, change the AI model's training objectives, or sell the company's data rights to a government.
Investors will have no way to stop him. The only check is the board, but the board is appointed by the CEO. Circular logic.
The Structural Contrarianism: What the Bulls Got Right
I am not arguing that super-voting shares are always malicious. The bulls have a point: founder-led companies often outperform their peers. Amazon, Google, and Meta all had strong founder control during their most innovative periods. The argument is that long-term vision requires insulation from quarterly earnings pressure.
But the context is different. AI is not search ads. The stakes are existential. The industry is already grappling with the potential for misuse, bias, and catastrophic failure. A governance structure that concentrates power in one person ignores the lessons of the last decade: centralization is a vulnerability.
Moreover, the super-voting structure may actually harm the IPO. Institutional investors are increasingly wary of multi-class structures. The Council of Institutional Investors has opposed such structures for years. Some index funds, like BlackRock, have policies limiting their investment in companies with unequal voting rights. If Anthropic faces a cold reception from large asset managers, it may have to lower its valuation or increase the float.
In the crypto world, we have a term for this: the liquidity premium. When a token has concentrated ownership, the market discounts it. The same principle applies to equities.
Takeaway: The Accountability Call
Anthropic's governance strategy is a bet against the very principles it claims to uphold. The company's safety rhetoric is hollow if the CEO can unilaterally change the rules.
When the CEO holds the keys to both the model and the governance, the only audit that matters is the one they refuse to commission.
The market will eventually price in this risk. The question is whether the IPO will succeed before the market realizes it.
Volatility is just liquidity leaving the room.
Technical Postscript
I want to add a layer of forensic detail for those who care about the numbers.
During my FTX ledger reconciliation, I manually traced the flow of $1.8 billion in missing funds. The pattern was simple: one entity controlled the wallets, the governance, and the accounting. There was no external check. The super-voting structure at Anthropic mirrors that centralization. The difference is that FTX was a fraud from the start. Anthropic is likely operating in good faith today. But the structure is the same.
I have also tested AI-generated audit tools against manual protocols. In 2024, I injected a logic flaw into a DeFi protocol during its fundraising phase. The automated scanners missed it. Human intuition caught it. The lesson: no algorithm can replace a skeptical human with a forensic mindset.
Anthropic's governance model relies on the same fallacy: that a single person can be trusted to act in everyone's interest. The data says otherwise.
The Broader Impact on AI Sector Investment Trends
If Anthropic's IPO succeeds despite the super-voting structure, we will see a wave of copycat filings. Every AI startup with a grand vision will demand the same terms. The result will be a sector where founder control is the norm, and investor rights are weakened.
This is not a prediction. It is a structural inevitability. The incentives are aligned for it. VCs want to protect their relationships with founders. Founders want control. The public market is the only party that can push back, but it rarely does.
The crypto market learned this lesson the hard way. DAOs with plutocratic governance structures failed. Those with quadratic voting or delegated proof-of-stake thrived. The same principles apply to corporate governance.
Conclusion: The Final Data Point
Anthropic's super-voting shares are a governance exploit. The market will treat them as a feature until it realizes they are a bug. By then, the damage will be done.
I will continue to monitor the filings. The on-chain data, when available, will tell the real story. Code doesn't lie. People do.