The artificial intelligence firm behind the Claude chatbot is preparing for a potential initial public offering that could value the company at a staggering $2 trillion.

However, once it hits the public markets, the company must confront a thorny issue: an external trust controls the majority of its board seats, and it remains to be seen whether this experimental governance model can withstand the intense scrutiny of public investors.

The company has established a "Long-Term Benefit Trust" (LTBT), a small group of advisers tasked with ensuring the company stays true to its mission of "benefiting humanity over the long term," even as commercial pressures mount. The trust holds no equity in Anthropic, yet wields immense power, with the authority to appoint or remove a majority of the board. Currently, four of Anthropic's seven directors are selected by the trust, including Netflix co-founder Reed Hastings and Novartis CEO Vasant Narasimhan.

The trust currently has three of its five potential seats filled, chaired by Neil Buddy Shah, CEO of the Clinton Health Access Initiative. Its other members include former Federal Reserve Chair Ben Bernanke and Richard Fontaine, CEO of the Center for a New American Security. Former California Supreme Court Justice Mariano-Florentino Cuéllar briefly joined the trust but departed after a few months to take on the role of Anthropic's head of global affairs.

According to several people familiar with the company's plans, Anthropic hopes this trust will set a benchmark for how future AI companies are governed. The goal is for the LTBT to become an industry-standard template, much like GAAP accounting rules—which also began as a voluntary standard proposed by private enterprises. One insider noted that inviting Bernanke to join in July was a deliberate step to give the trust more authority and institutional credibility.

Where to Begin with Its Influence

The trust's role is deeply involved: the company must notify the trust in advance of significant actions, such as releasing a new AI model. Members meet weekly and have sessions with company leadership every other week. They also sit in on regular board meetings and discuss major topics with the founders. For instance, during the launch of the cybersecurity model Mythos, the trust encouraged a limited release through the "Project Glasswing" initiative. It was also consulted during a dispute with the U.S. government over autonomous weapons.

Despite these extensive powers and access to information, the trust has so far largely played an advisory role. According to one person familiar with its operations, the trust has never drawn a red line or forced the company to make a major trade-off between profit and mission. In other words, this structure has not yet been truly tested—whether it can actually constrain company leadership when commercial and social goals diverge remains an open question.

Potential for Internal Conflict

Jesse Fried, a Harvard Law School professor specializing in corporate governance, argues that this structure has "conflict built into it." In a recent paper, he pointed out that Anthropic raises funds from profit-seeking investors while simultaneously allowing a small group of self-selected trustees to decide "how much profit to sacrifice in service of the mission"—a tension deeply embedded in the company's DNA. He also noted that OpenAI faced similar issues.

Other experts have also sounded warnings: while Anthropic's trust design is thoughtful, it has no precedent. The company is still losing money and will need to achieve long-term, sustainable profitability, at which point this mechanism will face its true stress test.

Elizabeth Pollman, a law professor at the University of Pennsylvania and a corporate governance expert, said it is nearly impossible to write every potential scenario into a contract to balance competing interests within a company. She specifically highlighted the intense competition in the AI field, both at the geopolitical and corporate levels, which makes such a balance even harder to achieve.

"The real challenge is whether this governance structure can function as envisioned over the long run, juggling dual or even multiple interests," she said.

Recent Trends and Comparisons

In recent years, Silicon Valley tech companies have trended away from traditional governance norms, making it harder for shareholders and the public to hold them accountable. Both Anthropic and OpenAI, the two major AI labs, have designed unconventional structures—rather than being managed by a traditional board that solely bears fiduciary duties, they have established self-selected "mission guardians."

OpenAI originally began as a non-profit, but in November 2023 it became a cautionary tale when its board attempted to fire CEO Sam Altman, only to lose the trust of investors and employees. Most of the directors were eventually replaced, and the company underwent a massive restructuring.

Anthropic's trust, in contrast, is considered lower-risk. According to sources, its structure includes an "emergency switch": trustees can be removed if 85% of shareholders vote to do so. This supermajority clause may change after the company goes public.

People close to Anthropic say that private investors in its multiple funding rounds were fully aware of this governance structure, and many of them explicitly cited "prioritizing safety" as one of their reasons for investing. However, one venture capitalist who supports Anthropic acknowledged that early investors accepted a basic premise: the company must ultimately become a commercial powerhouse to achieve its core mission.

"Investors at that time judged that capital power would ultimately prevail. No matter what you say, if you need massive amounts of money to buy computing power and compete, investors will naturally assume this is a business," the VC said.

Facing a Broader Investor Base

Now heading to the public markets, Anthropic will face a broader, and possibly more demanding, group of investors while profit pressures continue to mount.

Fried of Harvard wrote in his July paper: "Under the current structure, these directors have almost no personal stake in their decisions, and can even be required to ignore profits. OpenAI has already caused a stir. Anthropic's structure is slightly less risky, but it hasn't had an incident yet. Investors should carefully examine the arrangements of both companies—which may still be adjusted before their IPOs—and price their shares accordingly."