The boards governing Anthropic, OpenAI, and their peers face questions no governance framework was designed to answer. Their solutions will shape every other board's options.
This is not hyperbole. The governance decisions being made at frontier AI companies right now — about safety thresholds, deployment criteria, compute governance, and the structure of board authority itself — are establishing precedents that regulators will reference, that litigation will invoke, and that every downstream organization deploying AI will eventually have to reckon with. The hospital board deciding whether to adopt an AI diagnostic tool is operating in a governance landscape being actively shaped by what happens in the boardrooms of the companies building that tool.
A Tension with No Analog
Every industry has some version of the tension between growth and risk management. Pharmaceutical companies balance drug development speed against safety trial rigor. Financial institutions balance lending volume against default exposure. Energy companies balance extraction against environmental liability.
Frontier AI companies face a version of this tension that has no precise analog in any other sector: the capability they are building may be categorically different from anything that has existed before, and the people building it disagree, in good faith and with significant technical expertise, about exactly how different, how soon, and what that implies for how fast to move.
This is not a manageable risk calculus — it is a genuine epistemic problem. The board cannot resolve it by hiring better experts or commissioning more studies. It must govern under uncertainty that is structural, not temporary. And it must do so while the organization is competing with peers who may be making different bets about how much caution is warranted.
No governance framework was designed for this. Standard corporate governance assumes that boards can, in principle, understand the risks they are authorizing — that with enough expertise and diligence, the board can satisfy itself that management's risk assessment is reasonable. At frontier AI companies, that assumption breaks down. The risks are not fully knowable, the expertise required is concentrated in the very researchers the board is supposed to oversee, and the competitive pressure to move faster is constant.
The OpenAI Board Crisis as Governance Case Study
The events of November 2023 at OpenAI are frequently described as a leadership crisis or a personality conflict. They were neither, primarily. They were a governance failure — one that reveals exactly how inadequate conventional corporate governance structures are for the problems frontier AI companies face.
The structural problem was this: OpenAI had constructed a governance architecture in which a nonprofit board held ultimate authority over a capped-profit subsidiary, with a stated mission of ensuring that artificial general intelligence benefited humanity. The board had the legal authority to remove the CEO. What it lacked was any shared framework for when exercising that authority was warranted — what evidence would satisfy the threshold, what process would govern deliberation, and how the board would communicate its reasoning to the organization and the world.
When the board acted, it did so without that framework in place. The result was not just organizational chaos — it was a demonstration that formal authority without legitimate process is insufficient for governance at this level. The board had the power. It lacked the infrastructure to use that power in a way that produced a durable outcome. Within five days, the CEO was reinstated and the board composition changed substantially.
The lesson is not that nonprofit governance is wrong for AI companies. It is that any governance structure — nonprofit, PBC, or conventional corporation — requires explicit frameworks for the hardest decisions, not just the authority to make them.
Formal authority without legitimate process is not governance. It is the appearance of governance until the first genuine crisis — at which point the absence of real infrastructure becomes visible to everyone.
How Structure Changes Accountability: The PBC Model
Anthropic's choice to organize as a Public Benefit Corporation rather than a nonprofit or conventional C-corp is a meaningful governance decision, not merely a legal formality. PBC status creates a legal mandate to consider public benefit alongside profit — which means the board has a documented basis for decisions that prioritize safety over near-term revenue, and a legal framework that makes such decisions defensible to investors who might otherwise challenge them.
This matters practically. A conventional C-corp board that authorizes a pause in capability development for safety reasons faces potential shareholder challenge under the business judgment rule — why are we slowing growth? A PBC board making the same decision has a cleaner legal basis: this is within the scope of the public benefit mission the organization was incorporated to pursue.
The PBC structure does not solve the epistemic problem — the board still must make judgment calls about risks that are genuinely uncertain. But it changes the governance environment in which those calls are made, and it signals to the talent, investors, and regulators who interact with the company what the organization believes its accountability ultimately runs to.
Why Every Other Board Is Downstream of These Decisions
The governance choices frontier AI companies make are not just their own business. They establish the parameters within which every other organization deploys AI.
When a frontier AI company sets a safety policy — what content its models will and will not generate, what use cases it will and will not support, what monitoring it requires of API customers — it is making governance decisions on behalf of every hospital, university, financial institution, and nonprofit that deploys its technology. The hospital board approving an AI clinical decision support tool built on a foundation model is inheriting the governance choices made by the foundation model company's board, whether or not the hospital board knows it.
When frontier AI companies engage with regulators — in the EU under the AI Act, in the US through voluntary commitments, in the UK through the AI Safety Institute — the frameworks they help shape will constrain what regulators subsequently demand of AI users. The reporting requirements, audit standards, and liability frameworks that will govern AI use across sectors are being negotiated now, with frontier AI companies as primary interlocutors.
This is not an argument that frontier AI boards bear sole responsibility for the governance landscape. It is an argument that every other board should be paying attention to what they decide — because the governance options available to hospitals and universities and nonprofits in 2028 will have been significantly shaped by the choices being made in AI company boardrooms in 2025 and 2026. The boards that understand this will be better positioned to advocate for governance frameworks that serve their organizations' interests. The boards that treat it as someone else's problem will inherit the results of decisions they never participated in shaping.