Anthropic's Post-IPO Governance: A Detailed Analysis

Anthropic, an AI enterprise committed to ensuring AI remains beneficial and controllable for humanity, is embarking on its public market journey. However, the company's governance framework is notably intricate, diverging from typical structures seen in publicly traded entities. This design effectively limits the influence of general public shareholders on key corporate decisions, including board appointments.

Understanding Anthropic's Unique Governance Structure

Founded in 2021 with a core mission to ensure the long-term safety and benefit of artificial intelligence for humanity, Anthropic is now making its public debut. Yet, the company has implemented a highly complex governance system that significantly curtails the decision-making power of ordinary shareholders. This intricate setup ensures that the company's commitment to AI safety remains paramount, even at the expense of traditional investor control.

First, Anthropic operates as a Public Benefit Company, a legal designation that allows it to prioritize its public mission alongside profitability. This status empowers its board to make choices aligned with AI safety principles, safeguarding them from potential lawsuits by investors seeking purely profit-driven decisions. The current board, led by President Daniela Amodei and CEO Dario Amodei, also includes prominent figures like Netflix co-founder Reed Hastings and former Federal Reserve chair Ben Bernanke, among others.

Second, a crucial element of Anthropic's structure is the Long-Term Benefit Trust (LTBT). This independent body holds a special class of stock, granting it the authority to appoint and remove a majority of the company's board members, known as Class T directors. The LTBT, chaired by global health expert Neil Shah, and comprising distinguished individuals such as Ben Bernanke and Christopher Stubbs, is specifically tasked with upholding the company's mission to benefit humanity. Any vacancies on the LTBT are filled by its existing members, further solidifying its independence.

Third, the seven original founders of Anthropic collectively possess a unique class of stock that grants a majority of them 50.1% of the total shareholder voting power. This substantial influence enables them to elect three members to the board of directors. Only in the absence of a founder majority are their shares voted proportionally with common stock holders.

Finally, a mechanism exists to balance the LTBT's power: the board's nominating committee can veto the LTBT's selection or removal of a "Specified" director without cause. In such cases, the decision is then put to a vote among the company's employees. This intricate web of checks and balances creates a highly nuanced power dynamic within Anthropic.

The complexities of such a multifaceted governance model are difficult to fully foresee. Similar situations, such as the internal upheaval at OpenAI, where the board briefly ousted CEO Sam Altman due to concerns about transparency, highlight the potential for unexpected challenges. In OpenAI's case, employee loyalty ultimately led to Altman's reinstatement and a restructuring of governance. Anthropic's design appears to formalize a similar level of employee influence. This means that a fracture within the founders' unity could inadvertently empower common shareholders, while employee support for the board could override the LTBT's authority on certain director appointments. While the founders' control is set to expire ten years post-IPO or upon the reduction of their numbers to fewer than three, the LTBT's role will persist until 85% of common shareholders vote to terminate it or it voluntarily steps down with board approval. Ultimately, full control will eventually revert to common shareholders, ideally not as a consequence of an AI-induced crisis.