AI companies take different corporate forms and evolve over time. This report assessed whether function follows form, in the sense of whether certain business structures lend themselves to carrying out publicly minded missions more so than others. The case studies testing how various corporate vehicles walk the talk on their public missions suggest that corporate form has some bearing on how companies approach regulation, but little impact on safety and military policy. When competitive pressures and lucrative contracts are in play, financial interests prevail irrespective of the legal vehicle under which a company operates. In the competitive market that is AI, corporate structure alone cannot align companies with their mission. In fact, those very structures buckle under investor pressures, as is the case with OpenAI gradually abandoning its non-profit ethos to raise more capital.

When it comes to enforcing publicly oriented missions, recourse is limited. With the exception of preventing OpenAI from completely abandoning its original non-profit mission, there are very few instances of material gains. That is not to say efforts to activate corporate law levers are futile; as even the prospect of trouble can, and likely has, constrained the actions of AI companies. Perhaps counterintuitively, time-tested investor protection for publicly traded stock offers more impactful avenues to force disclosures. Should OpenAI and Anthropic proceed with their plan to go public, they, too, will be subject to increased transparency requirements. Accordingly, one important insight and avenue for future advocacy consists in activating disclosure requirements for publicly listed AI companies.

Three ways forward stem from this report. First, current corporate levers are underutilized. For example, OpenAI PBC’s board duty to exclusively consider the mission pursuant to its deal with the AGs opens the company to more litigation, especially as the company goes public and shareholders can form coalitions to gather the sufficient stock threshold to proceed with a derivative action on behalf of the PBC. That said, this avenue remains speculative as courts remain extremely reluctant to second-guess directors. Consequently, shareholder activism alone will not align companies with their mission. The limits of current legal vehicles call for new structures that actually have a forcing function on corporate behavior regarding mission compliance. PBCs stop short of that, as they allow but do not require companies to put mission above profits. They are meant to enable profit-driven companies to pursue social goals, not the other way around.

Third and perhaps more importantly, accountability requires robust regulation and public infrastructure. We need laws to move the needle towards publicly beneficial AI. Irrespective of their specific form—and without denying differences in degrees between various corporate vehicles—the self-interest of private companies cannot be meaningfully checked without laws that guarantee companies will factor in the costs of non-compliance. Corporate governance alone cannot ensure companies prioritize their mission over profits. Given the collective action problem that arises from competition among them (something key players recognized as early as 2019),346Amanda Askell, Miles Brundage & Gillian Hadfield, (<)em(>)The Role of Cooperation in Responsible AI Development(<)/em(>) (July 10, 2019) (unpublished manuscript), (<)a href='http://arxiv.org/abs/1907.04534'(>)http://arxiv.org/abs/1907.04534(<)/a(>). there is a fundamental mismatch between the incentives of AI companies jockeying for dominance in a capital-intensive market and the broader public interest. As Anthropic’s cofounder Jack Clark put it: “I am pretty skeptical of things that relate to corporate governance because I think the incentives of corporations are horrendously warped, including ours. […] It’s a real weird thing that this is not a government project.”347Dylan Matthews, (<)em(>)The $1 Billion Gamble to Ensure AI Doesn’t Destroy Humanity(<)/em(>), Vox (July 17, 2023), (<)a href='https://www.vox.com/future-perfect/23794855/anthropic-ai-openai-claude-2'(>)https://www.vox.com/future-perfect/23794855/anthropic-ai-openai-claude-2(<)/a(>); (<)em(>)see also(<)/em(>) Deep Ganguli et al., (<)em(>)Predictability and Surprise in Large Generative Models(<)/em(>), (<)em(>)in(<)/em(>) 2022 ACM Conf. on Fairness Accountability & Transparency 1747 (2022), (<)a href='http://arxiv.org/abs/2202.07785'(>)http://arxiv.org/abs/2202.07785(<)/a(>) (noting economic and prestige incentives to move forward even absent adequate assurances). Even OpenAI tried to secure public funding before abandoning its non-profit orientation.348The Ezra Klein Show, (<)em(>)supra(<)/em(>) note 69 (quoting Altman: “A little known fact, we tried to get the public sector to fund us before we went to the capped profit model. There was no interest.”). Policymakers need to step up with AI regulation that motivates companies to align their actions with the public interest, along with substantial investment in civic-minded public AI infrastructure.