This model is seen as a regulatory moat in the primary market, while raising concerns about shareholder dilution in the secondary market, highlighting a deeper conflict between capital logic and AI’s public responsibilities—and may become a new paradigm for top AI companies going public.Author and source: AIBase
Most eyes are on Anthropic’s revenue growth rate, model capabilities, and trillion-dollar valuation expectations as it races toward its IPO, yet few notice its governance structure, which diverges sharply from the typical Silicon Valley model. While most tech companies go public with dual-class AB shares to keep voting power in the founders’ hands, Anthropic has chosen neither path—instead, it has entrusted actual control of the company to a group of external trustees who hold no equity and receive no profits, including former Federal Reserve Chair Ben Bernanke. This design directly challenges a long-standing core assumption in capital markets: that public companies should prioritize shareholder returns.
The structure supporting this arrangement is a two-tier framework consisting of a Public Benefit Corporation (PBC) combined with a long-term independent trust. As a legal entity, the PBC remains a for-profit company capable of normal fundraising, issuing shares, listing on exchanges, and distributing dividends while participating in market competition. Its most fundamental change lies in redefining the fiduciary duties of the board: under its governance rules, management is no longer required to prioritize shareholder returns above all else and may legally factor in public safety and societal impact into decision-making. Above this, Anthropic has added a layer of a Long-Term Benefit Trust (LTBT), endowed with real authority and composed of externally selected experts who hold no equity in the company, receive no upside from listing or dividends, and are compensated only for their time. The trustees hold two core powers: first, the authority to replace a majority of board members; second, the requirement that the company must provide five days’ advance notice to the trust before engaging in significant non-routine transactions—such as substantial debt issuance, selling more than half of its equity, or transferring core AI models above a defined threshold. Many corporate law scholars who have analyzed these provisions conclude that this five-day notice period effectively functions as a soft veto: should the trust raise concerns about potential security risks in a proposed transaction, it can block its implementation through personnel changes and sustained pressure.
Looking at the global leading AI赛道, three distinctly different governance paths have emerged. OpenAI’s commercial operating entity is also a PBC, but its top-level structure is a nonprofit foundation that holds control, with a for-profit subsidiary handling commercialization. Anthropic’s entity remains a for-profit PBC, but it has separated control by transferring voting rights to an independent trust—the trust holds no equity, only voting power. xAI, founded by Musk, briefly registered as a PBC before abandoning it entirely and reverting to a conventional for-profit company, shedding its mission constraints and focusing all decisions squarely on commercial growth, with safety safeguards relying solely on corporate self-regulation. Anthropic occupies a middle ground: it retains access to market-based financing, avoiding the constraints imposed on OpenAI by its nonprofit parent, while leveraging an independent trust to make its safety mission a binding obligation, attempting to balance the freedom of commercial expansion with the public responsibility of AI technology.
Why are large model companies so focused on corporate responsibility? The core reason lies in the strong externalities of technological risks—using a standard corporate governance framework could easily pressure management to relax model permissions and deploy high-risk initiatives for short-term financial gains. Although the PBC legally allows the board to prioritize AI safety over profits, it struggles to prevent shareholders or founders from shifting direction under immense commercial incentives. Therefore, Anthropic has additionally established the LTBT, creating a dual constraint: the PBC is merely a legal vehicle; the real power—and the most contentious element—lies with a group of third-party trustees independent of shareholders and the founding team.
This architecture has spawned completely opposing views between the primary and secondary markets. The bull case in the primary market is clear: as a third-party gatekeeper, the trust avoids the common power struggles and internal friction between founders and investors seen in Silicon Valley, ensuring strategic continuity; in the context of globally tightening AI regulation, its inherently constrained governance structure serves as a compliance moat—capital cannot easily pressure the team to take technical risks for quarterly earnings, allowing slow-moving factors like R&D and security assessments to consistently receive resources. The secondary market, however, worries about practical realities: shareholder rights are structurally diluted; investors buying shares have little to no influence over transactions that could fundamentally alter the company’s trajectory, as control over major decisions rests with a group of trustees who are not accountable to shareholders. If the trust blocks a highly profitable merger or licensing deal on grounds of security, shareholders can only passively accept the outcome. Traditional due diligence can quantify revenue, gross margins, and competitive dynamics, but it cannot build valuation models around the subjective judgments of trustees—creating an inescapable paradox: the company possesses all the hallmarks of a high-growth prospect, yet the equity carries incomplete rights, and the market will ultimately discount the stock to reflect this rights deficiency. The primary market bets that long-term potential can absorb such a discount, while secondary market capital, operating under defined performance horizons, is naturally more cautious about this governance risk. This divergence in perception will shape pricing dynamics for a long time after listing.
Viewed over a longer time horizon, the significance of this matter extends beyond Anthropic’s own IPO success or failure. In the short term, the current high level of market enthusiasm for AI is sufficient to overshadow governance controversies, making the offering likely to proceed smoothly; in the medium term, as stock price volatility increases and major transactions materialize, debates over excessive trustee power and opaque rules will resurface repeatedly—should trustee decisions directly conflict with investor interests, shareholder lawsuits, institutional portfolio rebalancing, and valuation reassessments become foreseeable outcomes. Over the long term, if this mechanism can successfully balance commercial expansion with risk management, it has the potential to become a new blueprint for leading large model companies entering public capital markets. Having operated for a century, capital markets are built on the foundation of shareholder primacy—but the risk spillovers and societal impact of AI are unprecedented compared to internet or semiconductor industries. The old capital rules may not be suited to this new technology. Anthropic is using this IPO to test the waters on behalf of the entire industry; where this experiment ultimately leads will define how future giant AI companies should enter public markets.