Most companies have a simple goal: make money for shareholders. Anthropic, on its very first day, deliberately put a constraint on itself: it turned “making AI safe” into a legally mandated goal, and designed a mechanism that lets a group of non-shareholding outsiders ultimately control the board.

This piece walks through that unusual governance structure: what a Public Benefit Corporation (PBC) is, how the Long-Term Benefit Trust (LTBT) works, who does the gatekeeping, and where the design is most tested. If you want the full picture of the company first, start with What kind of company is Anthropic.

To set the frame in one line: Anthropic is using its corporate structure to make a bet—that “institutions” can hold the safety line at the critical moment better than “people” can.

First things first: what is a Public Benefit Corporation (PBC)?

A Public Benefit Corporation (PBC) is a special for-profit corporate form in the U.S. Its biggest difference from an ordinary company is that the charter explicitly requires the board, while pursuing shareholder returns, to also factor in a defined “public benefit.”

In other words, in an ordinary company a director who sacrifices profit for a mission could actually be sued by shareholders; but in a Public Benefit Corporation, a director “balancing the public interest” is authorized—even required—by the charter. For Anthropic, that public benefit written into the charter is “responsibly developing and maintaining advanced AI for the long-term benefit of humanity.”

This matters: it elevates “safety” from a corporate slogan to a goal at the legal level. Of course, the PBC form itself doesn’t automatically make a company safe; it merely opens a door, making “looking beyond the share price” a legitimate basis for decisions.

Worth noting: Anthropic isn’t the only AI company taking this route. Its biggest rival, OpenAI, adopted the public benefit corporation form for its for-profit entity after its 2025 restructuring, and observers often compare the two structures.

The truly unusual part: the Long-Term Benefit Trust (LTBT)

If it were just a public benefit corporation, Anthropic wouldn’t be one of a kind. What’s genuinely rare is the Long-Term Benefit Trust (LTBT), announced in 2023.

The trust’s members are a group of outside trustees who hold no shares in the company. Unlike ordinary directors or shareholders, they are nominally accountable not to the share price but to “the long-term interests of humanity.” Through a specially designed class of stock (which the company calls Class T), the trust gradually gains the power to appoint and replace some directors over time, with the eventual goal of being able to choose a majority of the board.

The problem this design tries to solve is direct: if AI is becoming powerful enough to shape the direction of society, then the company building it shouldn’t answer only to the highest bidder. Handing part of the power to appoint directors to a group with no financial stake is meant to give the “safety mission” a fulcrum that capital can’t easily buy away.

Three trustees are currently public: Clinton Health Access Initiative CEO Neil “Buddy” Shah, Center for a New American Security CEO Richard Fontaine, and former California Supreme Court Justice Mariano-Florentino Cuéllar. The trust was originally designed to have five seats; the names and status of the other two have not been fully disclosed by the company.

Is the trust already “live”?

A common question: it all sounds ideal, but does the mechanism actually have teeth, or is it just decoration?

Judging by how things have developed in 2026, the trust has actually started to act. In April 2026, Novartis CEO Vas Narasimhan was appointed to the board via the Long-Term Benefit Trust; after that, directors appointed by the trust now make up a majority of the board. That means the trust isn’t just a design on paper—it has exercised the core power it was given.

A caveat: this doesn’t mean the trust can direct day-to-day operations at will. Its power is concentrated at the level of “appointing and replacing directors”; the company’s strategy, products, and business decisions are still run by the management team and the board. The trust is more like an “ability to swap people out in the worst case” insurance policy than a day-to-day steering wheel.

As for exactly how much voting power Class T stock holds and under what conditions it can trigger replacing directors, the company has not fully disclosed—this remains one of the less transparent corners of the structure.

The gatekeepers: the board and the founding team

To understand a company’s governance, looking at “who sits at the table” is often the most direct approach.

The board currently includes CEO Dario Amodei and President Daniela Amodei—two internal founders—along with several outside directors: Netflix co-founder Reed Hastings; Chris Liddell, who led General Motors’ large IPO and served as Microsoft’s CFO (joined February 2026); Novartis CEO Vas Narasimhan, appointed by the trust (joined April 2026); and Yasmin Razavi. In recent years the board has clearly leaned toward “finding people with large-company governance and finance experience,” which is usually a signal that a company is preparing for greater scale—or even a potential public listing.

The founding team is another main thread. Anthropic was founded in 2021 by eight co-founders, all of whom came from OpenAI:

  • Dario Amodei (CEO): former OpenAI VP of Research, the company’s primary public spokesperson, who frequently represents the industry on policy and safety issues.
  • Daniela Amodei (President): Dario’s sister, former OpenAI VP of Operations, responsible for commercial operations and team building.
  • Jared Kaplan (Chief Science Officer): one of the founders of scaling laws research, who also serves as Responsible Scaling Officer, making the scientific calls behind the safety commitments.
  • Chris Olah: a leading figure in interpretability research, anchoring Anthropic’s research reputation in “understanding what’s inside the model.”
  • Jack Clark: former OpenAI policy lead, who drove early AI policy advocacy.
  • Tom Brown, Ben Mann, Sam McCandlish: core technical founders, with Tom Brown among the lead authors of the GPT-3 paper.

The roster itself explains Anthropic’s DNA: a group of people who once led research, safety, and policy at OpenAI, who left because they believed AI safety deserved to be taken more seriously, and who then wrote that conviction into the structure of a new company.

The biggest tension in this bet

However well the governance structure is described, the real test is always in reality. Anthropic’s core tension is that it’s doing two things that point in opposite directions at the same time.

On one side is commercialization at a pace rarely seen in history: within six months, the valuation jumped from about $183 billion, to $380 billion, all the way to nearly a trillion at $965 billion, and the company signed long-term compute contracts worth tens of billions of dollars each with cloud giants such as Amazon, Google, and Microsoft—deeply tied at the capital level. On the other side, it relies on the safety structure of the PBC, the Long-Term Benefit Trust, and the Responsible Scaling Policy to maintain a “safety-first” reputation publicly.

These two forces will eventually collide over certain concrete decisions. Outsiders usually watch three points:

  • The line on defense use: in early 2026, Anthropic and the U.S. Department of War (formerly the U.S. Department of Defense) publicly deadlocked over whether Claude could be used for certain purposes. The Department of War side has its national-security considerations, while Dario Amodei stated that he would not accept demands to remove the related safety protections. Supporters treat this as a positive case of the governance structure “holding the line,” but who is right remains unsettled.
  • Adjusting safety commitments: some media reports indicate that Anthropic relaxed certain stronger early safety commitments; the company responded with a new version of the Responsible Scaling Policy, emphasizing the addition of more transparency and outside review. These two accounts should be read side by side.
  • Restricted release of capabilities: for certain powerful features deemed higher-risk, Anthropic chose to limit who gets access rather than fully open or fully lock them. This “half-open” approach is itself an expression of governance judgment.

These three points are the real-world test of whether “institutions can actually outweigh revenue pressure.” Each is progressing in its own way, and none has reached a clear conclusion yet.

The parts not yet laid out

When discussing governance, the biggest trap is treating “design” as “outcome.” Here we honestly flag a few things that remain unclear:

  • Incomplete trustee roster: the Long-Term Benefit Trust was originally designed with five seats, but the public list has only three; the people and status of the other two have not been disclosed.
  • Opaque voting details: exactly how much voting power Class T stock holds and what conditions trigger replacing directors have not been fully disclosed by the company.
  • Undisclosed ownership ratios: the specific stakes held by founders such as Dario and Daniela, and the exact equity percentages of the major investors, have no official figures.
  • Some executive titles to be clarified: for example, the Chief Technology Officer (CTO) role has appeared under different names in early documents and later announcements, and the timing of the formal handover has not been fully explained officially.

These gaps don’t mean the structure is flawed, but they remind us: many claims about “who really controls Anthropic” still rest on estimation, not conclusions the company has confirmed.

Penchan’s take

Step back from this design and Anthropic is really challenging a Silicon Valley default: that a company should pursue maximizing shareholder value.

Its answer is that for a company that might build a technology affecting all of humanity, relying only on a founder’s promise that “I’ll do the right thing” isn’t reliable enough—because people can be bought, can leave, can change their minds. So it chose to write the mission into the charter and hand it to a non-shareholding trust as the last line of defense. This is a design philosophy of “don’t trust people, trust only institutions.”

But institutions have their limits too. For a company whose valuation is nearing a trillion dollars while being deeply tied to its biggest backers, the moment that really tests this structure will be the day interests and mission collide head-on. That day hasn’t come yet, so today’s praise or doubt is still only halftime commentary. For anyone trying to understand Anthropic, what’s really worth watching is whether this structure actually works each time the company faces a conflict. How many safety slogans it has shouted is, by comparison, secondary.

Further reading: What kind of company is Anthropic, Anthropic’s valuation and IPO, Anthropic’s risk list.