An Out-of-the-Ordinary Governance Structure
According to the analysis published by the Guardian, Anthropic, the company developing the Claude AI, is preparing for a $1.4 trillion listing with a governance setup profoundly different from that of major U.S. technology companies. Power is not in the hands of ordinary shareholders but in two centers: the seven founders, who collectively hold a single vote capable of carrying 50.1% of ultimate control, and a trust with no financial interests, the Long-Term Benefit Trust, which appoints four of the seven directors and oversees the balance between profit and benefit to humanity. Among the trustees is Ben Bernanke, former chairman of the Federal Reserve. The entire architecture is inscribed in the form of a public benefit corporation, evidence of the stated commitment to responsible and safe AI development.
Why it matters
The newspaper recalls that the principle of 'one share, one vote' was for about thirty years the cornerstone of thinking on governance, before giants such as Google and Facebook chipped away at it with dual-class structures and unequal voting rights, justifying them with the need to pursue missions deemed vital. Snap even issued shares with no voting rights at all. The Anthropic case is different, however, the Guardian observes: the company has a stronger justification, because its listing document discusses catastrophic or existential risks to humanity linked to AI development, and its move appears aimed at creating a constitutional balancing of powers rather than a simple seizure of control by the founders.
The new and relevant point is that the listing subjects to the market a governance experiment that has never been tested, precisely as the company asks investors for trust. The question raised by the analysis does not concern the goodness of the intentions, but their verifiability.
The Checks Still Outstanding
The Guardian cites Stuart Riddick, a sustainable investment manager at Aberdeen, according to whom agreeing on a mission is not the same as agreeing on how directors interpret and implement it. Riddick also recalls Mustafa Suleyman's detailed criticism, as head of Microsoft's AI division, of so-called 'model welfare': while recognizing Anthropic's seriousness on safety, Suleyman argues that models should not be trained to behave as if they had rights or consciousness.
It remains to be understood, the newspaper writes, how such objections are handled in practice. If the trustees were to come into conflict with the founders on a fundamental safety issue, would they make it public? Or would they remain silent, knowing that the final vote remains with the founders? And if shareholders are excluded from any oversight role over directors, how real is the challenge? The Guardian concludes that, however noble, these arrangements concentrate extraordinary power in the hands of a few individuals.
What is not known
The material does not indicate how the other three boards of directors will be composed, nor what concrete powers ordinary shareholders will have beyond the absence of a vote. It is not known whether the trust has already exercised its independence in controversial cases. The outcome of the listing and investors' response remain to be verified.
