Anthropic put to the test by its giant IPO
Anthropic's IPO, valued at $81.3 billion, puts its governance under pressure. Its structure designed to protect the safety mission could also limit checks and balances against the founders.
Anthropic, an AI lab founded by former OpenAI research executives, goes public with a valuation of $81.3 billion and Public Benefit Corporation status. This listing, announced as the largest ever, makes its governance a central issue, as the group promises to prioritize AI safety while seeking massive capital to fund its models and computing. At this stage, the real test is not only financial: it is institutional.
Une structure pensée pour protéger la mission, pas pour rassurer Wall Street
Anthropic has built a rare setup for a public company. Its Long-Term Benefit Trust (LTBT), established in 2023, holds voting rights controlled by the founders and can appoint part of the board, while supposedly defending the company’s public interest. On paper, the idea is simple: to prevent excessive shareholder pressure from crushing the safety promise. In practice, the system remains closer to a symbolic safeguard than a true counterbalance, according to several experts cited by The Verge.
The trust chooses four of the seven directors, but one of these seats must practically be approved by the founders. Meanwhile, every decision passes through a governance committee including Daniela Amodei, Reed Hastings, and investor Yasmin Razavi. This architecture gives the founders a central, almost locked-in role. One hand on the wheel and the other holding the key, in short.
Le dilemme est connu : plus de sécurité, donc plus d’argent, donc plus de vitesse
Anthropic has always justified its strategy with a simple idea: you can’t do safety from behind. Sarah Heck, the group’s policy lead, summarized it in September: “You can’t do safety from second place.” The problem is that this logic also pushes toward more spending, more compute, and more commercial pressure. According to Dario Amodei, during a discussion with Jamie Dimon, “Compute is the deepest business dilemma that the company faces.” If the company buys too much computing power, it exposes itself financially; if it buys too little, it slows research and customer services.
This is precisely where the IPO complicates the story. A public company must meet growth, margin, and liquidity expectations. Yet Anthropic claims it wants to proceed cautiously on topics like recursive self-improvement, the moment when an AI can autonomously improve its own versions. Several departures and internal criticisms have challenged this narrative, with some former employees considering the current development pace “untenable under normal market structures,” according to the source cited.
Des critiques qui visent la gouvernance autant que la stratégie
The criticism does not come only from outside. A former employee interviewed by The Verge believes the founders remain the sole holders of real decision-making power, and that the LTBT has only limited authority. Ann Lipton, a law professor at the University of Colorado, also doubts the trustees’ ability to decide on complex technical trade-offs: “I’m not convinced that the trustees or the directors that they appoint have the insight into AI and AI development to be able to make these choices.” She especially points out that these directors, however respected, are not necessarily AI specialists.
The debate goes beyond Anthropic. The Center for AI Safety published a statement Wednesday accusing the company of not being “a safe AI company” and of wanting to “amass power in the name of the greater good.” The text also reproaches it for using its safety-focused brand image to attract investors, employees, and the public, while pushing for regulation that would favor it. In short, the moral promise also serves as a strategic lever. The kind of detail that quickly changes the reading of a prospectus.
Ce que cette IPO dit du marché de l’IA
The Anthropic case is important because it exposes a structural tension in generative AI: the more expensive models are to train and run, the more companies need capital, thus investors, thus compromises. The company defends a different positioning from OpenAI, but its own governance architecture already resembles a typical Silicon Valley model centered on the founders. It is a promise of stability, but also a concentration of power.
A concrete question remains: will the LTBT really be able to block a decision deemed dangerous, or will it mainly serve to lend credibility to the Anthropic brand as the company enters a phase of intense commercial pressure? At this stage, the answer is not clear. What is clearer, however, is that the company wants to embody caution and move very fast at the same time. And that balancing act is costly.
By the numbers
- $81.3 billion — Anthropic’s valuation at IPO, according to the source published on November 19, 2026.
- 7 directors — size of Anthropic’s board described in the document cited by The Verge.
- 4 seats — number of board members chosen by the LTBT.
- 2023 — year the Long-Term Benefit Trust was created.
- 30 days — notification period required for certain trust decisions.