Anthropic's $2 Trillion IPO Puts Its Unusual Governance Experiment Under the Microscope
When Anthropic eventually floats, public investors won't just be buying into an AI lab. They'll be buying into a governance structure that hands meaningful control to a small group of external trustees who hold no shares, answer to no shareholders, and can't be easily shifted. That's a pitch that works fine in a Series D round. Whether it flies on public markets is a different question.
The company's Long-Term Benefit Trust sits at the top of the governance stack. It has the right to appoint or remove a majority of Anthropic's board, and it has exercised that power: four of the company's seven directors are LTBT picks, including Netflix co-founder Reed Hastings and Novartis CEO Vas Narasimhan. The trust itself currently has three members: chair Neil Buddy Shah (head of the Clinton Health Access Initiative), former Fed chair Ben Bernanke, and Richard Fontaine of the Center for a New American Security. A fourth trustee, former California Supreme Court justice Tino Cuéllar, left after a few months to become Anthropic's chief global affairs officer, which tells its own story.
The trust holds no equity. It exists, at least in theory, to keep Anthropic honest about its stated mission of developing AI for humanity's long-term benefit rather than just turning a profit.
People close to Anthropic describe the LTBT as something of a blueprint exercise. The ambition, they say, is to create an industry standard for AI governance, something analogous to GAAP accounting principles, which started as voluntary private-sector frameworks before becoming entrenched norms. Bernanke's appointment in July is seen internally as a signal that the trust is meant to become a genuinely institutional body rather than a token ethics committee.
In practice, the trustees get advance notice of major company decisions, including new model launches. They meet weekly among themselves and with Anthropic's leadership roughly every other fortnight. They attend board meetings and hold direct discussions with founders. They apparently weighed in on the rollout of the Mythos cybersecurity model and on Anthropic's position regarding automated weapons contracts with the US government.
But here's the uncomfortable part. Despite all that access, the trust hasn't actually forced any hard choices. It has operated more as an advisory body than a genuine check. No red lines have been drawn. No significant trade-off between commercial goals and the public interest has been demanded. The structure, as Harvard Law professor Jesse Fried put it plainly in a recent paper, hard-wires a "deep and potentially unmanageable tension" into the company's DNA. Anthropic raises money from profit-seeking investors, then lets a self-appointed group decide how much profit is acceptable to sacrifice in service of a broader mission. That's not a problem when everyone's getting along. It becomes a very real problem when they're not.
Elizabeth Pollman, a corporate governance expert at the University of Pennsylvania, is similarly cautious. Contracting perfectly for every scenario where competing interests collide is, she says, essentially impossible. Throw in the geopolitical dimensions of AI competition and the commercial arms race between labs, and the balancing act gets considerably harder. "Will this governance structure work in the way intended, serving dual or more interests over time?" she asked. That question remains genuinely open.
Anthropologists of tech governance will recognise a pattern here. Silicon Valley has spent years drifting away from conventional accountability structures, giving founders outsized control through dual-class shares, mission-aligned charters, and various other mechanisms that make it harder for ordinary shareholders to exert pressure. Anthropic and OpenAI represent the most extreme end of that drift.
OpenAI's November 2023 board crisis, when directors tried to oust Sam Altman and promptly collapsed under pressure from investors and staff, remains the canonical warning about what happens when governance idealism meets commercial reality. Most of that board is gone. The company is now restructuring.
Anthropics structure has a safety valve OpenAI lacked: trustees can be removed with 85 percent of shareholder voting power. That's a high bar, but it exists. Whether the supermajority threshold survives IPO negotiations is another matter.
Venture capitalists who backed Anthropic early apparently did so knowing the structure was unusual and, at least partly, because of it. Safety-focused governance was part of the pitch. But there was also, one investor admitted, a quiet assumption that commercial logic would ultimately prevail. "Investors assume there will be a business," they said. That's not the same as assuming the mission stays intact.
Going public changes the pressure dynamics significantly. Public market investors are, on average, less patient and less ideologically invested than early-stage VCs. An unprofitable AI company asking them to trust a small group of unelected trustees with majority board control is a genuinely novel ask. Fried's conclusion was blunt: investors should scrutinise these arrangements carefully and price shares accordingly. Given that Anthropic could be valued at up to $2 trillion, there's a lot riding on how that scrutiny goes.