Jesse M. Fried, the esteemed William Nelson Cromwell Professor of Law at Harvard Law School, and Idan Reiter, an S.J.D. candidate at Harvard Law School, have published a groundbreaking paper that critically examines the corporate governance structures of leading artificial intelligence companies, namely OpenAI and Anthropic. Their research, detailed in the paper "AI Corporate Governance and Ben & Jerry’s Risk," uncovers a significant and potentially destabilizing inherent conflict within these organizations. This conflict arises from the unique arrangement where entities funded by profit-driven investors also empower self-appointed individuals to override investor decisions, influencing the extent to which profits might be sacrificed for the sake of ensuring AI benefits humanity. This fundamental tension, deeply embedded in the corporate DNA of these firms, presents a complex governance challenge with far-reaching implications.
The concept of "self-appointed mission guardians" employed by OpenAI and Anthropic echoes a past experiment at Unilever’s subsidiary, Ben & Jerry’s. This earlier instance, as detailed by Fried and Reiter, resulted in what they term "double trouble"—a scenario where the guardians not only negatively impacted investors but also inadvertently achieved the opposite of their intended mission. This historical parallel serves as a stark warning, highlighting the inherent risks associated with instituting such guardians. The authors suggest that this past failure may have influenced Anthropic’s design, which includes a crucial "kill switch" allowing a super-majority of investors to remove its guardians, a feature absent in OpenAI’s current structure.
The Peril of "Doing Too Much": Guardians Beyond Investor Control
The prevailing critique of mission guardians typically centers on the concern that they might "do too little." This perspective posits that without direct incentives to advance their mission and facing persistent pressure from founders, investors, and equity-compensated employees, these guardians would inevitably subordinate public benefit objectives to profit maximization. Fried and Reiter acknowledge this possibility but argue that a more significant and perhaps more dangerous risk lies in the guardians’ potential to "do too much."
Their analysis suggests that a lack of accountability to investors and the absence of incentives that align with investor interests can lead guardians to actions that not only harm investors but also undermine the very mission they are tasked to protect. The researchers point to two prominent cases—Ben & Jerry’s and OpenAI—where guardians, by overstepping their bounds, have endangered or harmed investors while simultaneously achieving outcomes antithetical to their stated mission. This leads to the unsettling conclusion that in certain circumstances, these guardians can be not only ineffective but actively detrimental.
The Ben & Jerry’s Precedent: A Case Study in "Double Trouble"
The history of Ben & Jerry’s offers a compelling, albeit cautionary, tale. In 2000, when Unilever acquired the iconic ice cream brand, a critical agreement was made to retain self-perpetuating independent directors. These directors were vested with the authority to override Unilever’s decisions in order to safeguard Ben & Jerry’s "social mission" and "brand integrity." For approximately two decades, disputes between Unilever and these independent directors were managed discreetly.
However, this delicate balance was dramatically disrupted in July 2021. The independent directors announced their intention not to renew the license held by Ben & Jerry’s Israeli licensee, a decision made against Unilever’s explicit objections. This move ignited a protracted and acrimonious battle. The ensuing conflict saw a series of retaliatory boycotts, divestments by several U.S. states, intervention from an activist investor, and multiple legal skirmishes between the directors and Unilever. The fallout was significant: Unilever’s CEO resigned, and the company reportedly suffered billions in market value depreciation, a sum substantially exceeding the total worth of Ben & Jerry’s itself.
Beyond the substantial financial repercussions for Unilever, the episode revealed a striking failure of the guardians’ mission. The directors had deemed continued sales in Israel inconsistent with their mission. Yet, their actions ultimately led to a situation where Unilever, in 2022, successfully overrode the directors. The Israeli licensee was granted the necessary provisions to continue selling Ben & Jerry’s ice cream in Israel and its controlled territories in perpetuity. This outcome directly contradicted the guardians’ stated intent.
Furthermore, in a move designed to insulate its remaining ice cream businesses from future interference, Unilever initiated a spin-off of its ice cream division in 2025. This strategic maneuver effectively severed the ability of Ben & Jerry’s guardians to impose further costs on Unilever. The researchers have termed this phenomenon "Ben & Jerry’s risk," referring to the potential for such a "double-trouble meltdown" where guardians inflict harm on investors and subvert their own mission.
OpenAI’s Meltdown: A High-Stakes Governance Crisis
The governance crisis that unfolded at OpenAI in November 2023 serves as another critical example of "Ben & Jerry’s risk." Similar to Ben & Jerry’s, OpenAI’s structure involved a juxtaposition of self-appointed mission guardians—the directors of the non-profit OpenAI, Inc.—and investors in its for-profit subsidiary, an LLC. The crisis erupted when the non-profit board abruptly fired CEO Sam Altman, citing, in part, safety-related concerns.
This decision nearly led to the collapse of the LLC and the decimation of its investors. The situation escalated rapidly as approximately 700 of OpenAI’s 770 employees threatened to resign and join Microsoft, a key investor. Faced with this overwhelming exodus, the board was compelled to reverse its decision. Altman was reinstated, the board members who initiated the firing faced pressure to resign, and reportedly, some of the most safety-focused researchers, including Mira Murati and Ilya Sutskever, ultimately departed to establish competing AI ventures.
The consequences for OpenAI’s investors were severe, with their stakes facing near-total loss. Paradoxically, the incident may have also compromised OpenAI’s safety objectives. By ousting key safety advocates and potentially scattering expertise, the move could have rendered OpenAI itself less safe—the very outcome the guardians purportedly sought to prevent.
Fried and Reiter’s analysis indicates that OpenAI’s subsequent restructuring in 2025 has done little to mitigate these risks. The for-profit arm has been reorganized as a Delaware public benefit corporation (OpenAI Group PBC). However, it remains under the significant control of the non-profit entity, now renamed the OpenAI Foundation. The Foundation retains the power to appoint every director to the PBC’s board, possesses veto rights over major transactions, and, through its Safety and Security Committee, can block any "PBC actions relating to safety and security." The Foundation has, in fact, appointed all but one of its own directors to the PBC’s board. Crucially, the fiduciary duties of the PBC directors do not constrain these guardians. Instead, they mandate that directors disregard investors entirely on safety and security matters and otherwise permit the subordination of profits to the mission.
Anthropic’s Safeguard: The "Kill Switch" as a Deterrent
In contrast to the governance struggles at Ben & Jerry’s and OpenAI, Anthropic’s corporate structure appears designed to mitigate the risk of a "Ben & Jerry’s-style meltdown." Like OpenAI, Anthropic features a controlling mission entity, the Anthropic Long-Term Benefit Trust, overseeing a public benefit corporation (Anthropic PBC). However, the key distinction lies in the relative alignment of Anthropic’s guardians with investors and their reduced power over them.
The most significant innovation in Anthropic’s governance is the inclusion of a "kill switch." This mechanism empowers a super-majority of Anthropic’s stockholders to terminate the Anthropic Long-Term Benefit Trust and remove the directors appointed by it to the PBC’s board. This provision effectively renders Anthropic’s guardians only "partly insulated" from investor oversight, unlike their counterparts at OpenAI, who are far more insulated.
Fried and Reiter anticipate that this investor oversight mechanism will serve as a powerful deterrent, discouraging Anthropic’s guardians from emulating the actions of those at Ben & Jerry’s and OpenAI, which led to "doing too much." Should these guardians nonetheless act in ways detrimental to investor interests, the kill switch provides a direct recourse for investors to remove them.
Implications for Future AI Governance
The experiences at Ben & Jerry’s and OpenAI, as the only two firms to date that have implemented fully insulated guardians, offer critical lessons for the burgeoning AI industry. Fried and Reiter conclude that future firms establishing similar governance structures would be wise to adopt Anthropic’s model, incorporating a "kill switch." This mechanism provides a vital safeguard, ensuring that the pursuit of ambitious AI missions does not come at the expense of investor rights and the financial stability of these critical technological enterprises. The authors’ research underscores the necessity of robust and accountable governance frameworks as the AI sector continues its rapid expansion and integration into the global economy.
The full paper, offering a comprehensive analysis of these complex governance dynamics, is available for review.
