Anthropic, founded in 2021 to keep AI under human control and ensure long‑term benefits, is nearing an initial public offering. Yet the company’s structure will keep most public shareholders out of the decision‑making process. Silicon Valley has long used complex share classes that let founders such as Elon Musk or Mark Zuckerberg steer public companies without owning a majority, but Anthropic’s focus on AI safety has produced an especially intricate governance model that limits the influence of ordinary investors, even over board appointments.
Anthropic is organized as a public benefit corporation, a for‑profit entity that can pursue a public mission alongside profit. This legal status shields the board’s AI‑safety decisions from lawsuits by investors who might argue that the choices are not purely profit‑driven. The current board is chaired by president Daniela Amodei and includes CEO Dario Amodei, her brother, Netflix co‑founder Reed Hastings, former Microsoft and GM CFO Christopher Liddell, Novartis chief executive Vas Narasimhan, and Spark Capital partner Yasmin Razavi. One board seat remains vacant.
A second layer of control rests with a Long‑Term Benefit Trust (LTBT), an independent panel tasked with keeping the company focused on its humanity‑benefiting mission. The trust holds a class of stock that can elect and remove a majority of the board, known as Class T directors. Neil Shah, a global‑health expert who formerly led GiveWell, chairs the LTBT, which also includes former Federal Reserve chair Ben Bernanke, national‑security specialist Richard Fontaine, and Harvard physicist Christopher Stubbs. The trust also has one open seat, to be filled by its existing members.
The seven founders together own a special class of shares that grants them 50.1 percent of the voting power, enough to elect three board members. If the founders cannot reach a majority, their votes are counted proportionally with common‑stock holders.
The board can also check the LTBT. When the trust nominates a “Specified” director or seeks to remove one without cause, the board’s nominating committee may refuse to approve the move and instead refer the decision to a vote of Anthropic’s employees.
These overlapping mechanisms make the power balance difficult to forecast. A comparable situation unfolded at OpenAI when its board briefly dismissed CEO Sam Altman over concerns about candor. Employee loyalty forced the board to reverse the decision and prompted a new governance structure. Anthropic’s design appears to formalize that kind of employee influence. If the founders’ unity cracks, ordinary shareholders could gain more sway than expected, while employees could override the LTBT on board seats if they side with the board over the appointed experts.
Founders’ control is not permanent. Ten years after the IPO, the founders may voluntarily relinquish their authority, or it will end automatically if fewer than three founders remain. At that point the LTBT will stay in place until 85 percent of common shareholders vote to dissolve it, or it steps down with board approval. Only when neither the founders nor the trust exist will common shareholders obtain full control, assuming the company avoids any AI‑driven disaster.