Anthropic's trust experiment: the real story behind Cuéllar's appointment

SHOULD A HANDFUL of men control the world's most powerful technology? That question has been around since the first large language models appeared. Anthropic's latest move suggests it has decided the answer is "yes, but with oversight by people who never owned a share of the company."

In January this year, the AI firm appointed Mariano-Florentino Cuéllar to its Long-Term Benefit Trust, the independent governance body that selects members of Anthropic's board and advises its leadership. Mr Cuéllar, a former justice of California's Supreme Court and current president of the Carnegie Endowment for International Peace, brings experience in law, public institutions and cross-border policy. He replaces two founding trustees, Kanika Bahl and Zachary Robinson, who built the trust from scratch after it was unveiled in 2023. The appointment was followed by another of a different register: last month the trust added Ben Bernanke, the former Federal Reserve chairman and Nobel laureate.

The headline describing Mr Cuéllar as Anthropic's first "global affairs chief" is incorrect. That role belongs to Michael Sellitto, Anthropic's head of global affairs and a former National Security Council cybersecurity adviser. Mr Cuéllar's seat is on a five-person body of financially disinterested trustees. They hold no equity in Anthropic, share in no profits and are compensated only for their time. Their authority, however, is not advisory in name only: the trust can appoint up to three of the company's five board directors, and was designed to reach majority board control within four years of establishment.

The structural point is worth pausing over. Anthropic, a Delaware public benefit corporation, has built a governance mechanism that attempts to insulate decisions about frontier AI from the ordinary pressures of shareholder capitalism. The trust was written into the company's Series A investment documents, a rarity for a venture-backed startup. Its members hold special Class T common stock that grants board-election power without economic stake. New trustees are chosen by existing ones, in consultation with the company. The aim, as Anthropic's co-founder Dario Amodei has framed it, is to ensure that decisions about the development of powerful AI systems remain grounded in the broader public interest, even when short-term commercial incentives pull in another direction.

To be sure, the experiment has been tentative. By late 2025 the trust had filled only one of its three available board seats, according to analysis on the Longterm Wiki, a research site tracking AI governance mechanisms. That slow start left open the question of whether the trust would prove a decorative feature or a functional brake. The appointments of Mr Cuéllar and Mr Bernanke, together with Richard Fontaine of the Centre for a New American Security and Neil Buddy Shah, who chairs the trust, suggest it is moving toward the latter. The roster now reads like a cabinet of former regulators, judges and security architects rather than a collection of effective-altruism volunteers. The signal is unmistakable: as the company scales, so too is the seriousness of its oversight.

The timing is not coincidental. In May Anthropic closed a $65 billion Series H funding round at a $965 billion post-money valuation, its run-rate revenue having crossed $47 billion. Three weeks later, on June 1st, the company confidentially filed a draft S-1 registration statement with the Securities and Exchange Commission, the first major AI lab to formally begin an IPO process. A public listing would transform the incentives surrounding the firm almost overnight. Public shareholders care about returns, quarterly earnings and share-price momentum. The very pressures the trust was designed to buffer against.

That is the real question beneath the personnel announcement. Can an independent trust with no financial stake resist a board and a management team that will soon answer to a stock market demanding growth, margin expansion and deployment speed? The academic literature has a name for the phenomenon: amoral drift. Professors Oliver Hart and Luigi Zingales at Harvard describe it as the slow erosion of prosocial corporate missions under market pressure. Their analysis was written with OpenAI in mind-the drama of Sam Altman's firing and reinstatement showed how quickly investor-supplier alliances can override a board's stated commitments. A chapter in the Harvard Law Review in 2025 extended the concern to Anthropic, arguing that even weakened shareholder rights may not be enough to prevent mission drift when the surrounding constituency exerts its own gravitational pull.

The trouble is not that the trust has no authority. It is that authority without enforcement teeth is largely ceremonial. The trust's governing documents can be amended by a stockholder supermajority, creating a potential override mechanism. The trust does not control the CEO. It does not set pricing, compute strategy, or product roadmaps. Its stated role is to intervene in extreme cases-ensuring that leadership evaluates future models for catastrophic risks, for instance, or that they maintain nation-state-level security rather than racing to market at any cost. These are precisely the moments when commercial pressure would be strongest, and when a financially disinterested trustee might find himself or herself isolated.

There is also the political dimension. Anthropic has made a deliberate choice to resist the prevailing style of Silicon Valley lobbying. The company declined to accept Pentagon contract language last April, prompting the defence department to label it a "supply chain risk". It was hit with export controls on its Mythos model after a jailbreak was discovered. Mr Amodei has been called a "liar" and an "ideological lunatic" by senior officials in the Trump administration, and Mr Trump himself has described the firm as staffed by "leftwing nut jobs". The company's response has been to double down on its safety stance, donating $20 million to a bipartisan AI-policy non-profit and arguing publicly that the government is "not doing enough to regulate" AI. This posture has won loyalty among its engineers and some enterprise customers. It has also made the firm a target.

For investors in the forthcoming IPO, the trust's credibility will be a governance question on the prospectus, not merely a mission statement on the website. If the trust proves capable of constraining decisions that prioritise speed over safety, it is a feature: it reduces tail risk from regulatory backlash, catastrophic failure, or loss of institutional trust. If it proves incapable, it is a liability: a mechanism that raises expectations without delivering them, inviting activist scrutiny or shareholder litigation when things go wrong.

The composition of the current trust is a step in the right direction. Mr Cuéllar's background spans three presidential administrations, a California Supreme Court term and leadership of a major international-policy institution. Mr Bernanke brings eight years at the Federal Reserve and deep expertise in systemic economic risk. Mr Fontaine has long run one of Washington's most influential defence think-tanks. The appointment of people who have held office in actual institutions, rather than only advised from outside them, raises the trust's credibility with regulators, governments and public-market investors.

But credibility is not enforcement. The deeper test will come when the trust is asked to say no. Will it push back on a model release that could accelerate cyber capabilities? Will it resist pricing pressure that might push Claude into less safe but more lucrative applications? Will it hold firm when a stock price is falling and the board wants to cut safety research to improve margins? The answer to those questions will determine whether the Long-Term Benefit Trust is a genuine innovation in corporate governance or an elaborate piece of theatre.

The Luddites were wrong about machines in the long run, but not about the pain of transition. The question for Anthropic is whether it has built institutions durable enough to manage that transition without losing its nerve. The trust's new members would suggest it thinks it has. The market, soon enough, will find out.