menu_open Columnists
We use cookies to provide some features and experiences in QOSHE

More information  .  Close

Who controls Anthropic? 10 parts of its unusual governance

4 0
yesterday

Get Quartz in your inbox

Free daily briefing on global business news.

Privacy and Cookie Settings

Who controls Anthropic? 10 parts of its unusual governance

Anthropic's ownership structure layers a Delaware public benefit charter, an independent Long-Term Benefit Trust and minority investors Amazon and Google — here is how each piece works

Anthropic has raised billions of dollars from Amazon $AMZN and Google $GOOGL. It has also given a group of people with no financial stake in the business the right to elect part of its board. Those two facts pull in different directions. The company's governance is built around managing that tension.

Anthropic was founded in 2021 by former OpenAI employees. Among them were siblings Dario Amodei, the company's CEO, and Daniela Amodei, its president. The company develops large AI models. That work requires vast amounts of computing power, and computing power requires capital. That need for money shapes how the company is governed.

The Anthropic ownership structure has three main layers. The first is its legal form. Anthropic is incorporated in Delaware as a public benefit corporation. That legal form requires its directors to balance shareholder returns against a stated public benefit. The second layer is the Long-Term Benefit Trust, which Anthropic announced in September 2023. The trust holds a special class of shares called Class T stock. Anthropic has said that stock lets the trust elect a portion of the board and, over time, a majority of directors. The third layer is outside capital. Amazon and Google have both invested as minority shareholders. Neither has board control.

Each layer answers a different question. The charter defines what the company is meant to serve. The trust gains a growing role in choosing the directors who interpret that purpose. The investors supply the money that building frontier AI models consumes.

The arrangement matters beyond Anthropic. Control of an AI developer shapes how it weighs commercial pressure against safety and long-term risk. Those trade-offs can surface in decisions about when to release a model, how to price it and which partners to work with. Investors, customers, regulators and competitors have reason to know where authority sits.

The design is not locked in place. Anthropic has said stockholders can alter the trust arrangement with a large supermajority vote. That provision means the trust's role can be revised, though only with broad stockholder agreement.

The 10 slides that follow take each piece in turn. They cover the public benefit charter and its purpose, the trust and its Class T stock, and how its board power phases in. They also cover who can serve as a trustee, the roles of Amazon and Google, the contrast with OpenAI and the open questions.

Anthropic is a Delaware public benefit corporation

Anthropic is incorporated in Delaware as a public benefit corporation, or PBC. That legal form requires its directors to weigh a stated public benefit alongside the financial returns owed to shareholders.

Delaware added the public benefit corporation to its corporate law in 2013. By adopting that form, Anthropic wrote a public mission into its certificate of incorporation. The mission does not sit in a separate statement outside the company's legal documents.

A PBC is still a for-profit company. It can issue shares, raise money from investors, sell products and earn profits. It is not a charity, and the label does not make it exempt from taxes. What changes is the set of obligations its board carries.

In a conventional Delaware corporation, directors answer chiefly to shareholders. Their decisions are judged largely by whether they serve shareholder value. A PBC adds more interests to that calculation.

Under Delaware law, a PBC's directors must balance three interests when they manage the company. The first is the financial interest of stockholders. The second is the best interests of people materially affected by the company's conduct. The third is the specific public benefit named in the certificate of incorporation.

That balancing duty gives a board legal room to weigh interests beyond share value. It does not tell directors which way to decide in any particular case. The statute asks for a balance, not a fixed ranking of priorities.

Delaware also requires a PBC to report to its stockholders at least every two years on how it is pursuing its public benefit. Enforcement of the balancing duty rests mainly with the company's own stockholders rather than with the public at large. The form is a governance tool built into corporate law, not a regulator that inspects the company from outside.

For Anthropic, the charter defines what the board must weigh. It does not settle who sits on that board or who chooses those directors. The PBC form leaves that question to other parts of the company's structure.

What Anthropic's public benefit purpose commits it to

Anthropic's stated public benefit is the responsible development and maintenance of advanced AI for the long-term benefit of humanity. That sentence is the goal its directors must weigh alongside returns to shareholders.

The phrasing is compact, and each term narrows or widens the commitment. "Development" covers building new AI systems. "Maintenance" extends the purpose past launch, to systems already in use. "Responsible" applies to both stages.

"Advanced AI" limits the scope to the frontier technology the company builds, rather than to technology in general. "Long-term" points the purpose at outcomes beyond the next product cycle or funding round. "Humanity" names a beneficiary far wider than Anthropic's customers, employees or investors.

The purpose does not name a product, a market or a revenue goal. It names an effect on people and a time frame for judging that effect.

The written purpose carries legal........

© Quartz