In AI corporate governance and Ben & Jerry’s risk, authors Jesse Fried and Idan Reiter, both of Harvard Law School, examine the governance arrangements at OpenAI and Anthropic through the lens of an earlier corporate experiment involving Ben & Jerry's, the ice cream brand formerly owned by Unilever.
The paper argues that all three organizations share an unusual governance feature: investors provide capital but do not have ultimate authority over how management balances commercial objectives against a broader social mission.
Instead, that authority sits with what the authors describe as ‘self-appointed mission guardians’, individuals empowered to determine how much profit should be sacrificed in pursuit of goals such as AI safety or social activism.
For corporate governance professionals, the paper raises wider questions about accountability, fiduciary oversight and whether boards insulated from investors can sustainably balance mission and value creation.
A built-in governance conflict
According to Fried and Reiter, the governance structures at OpenAI and Anthropic create a ‘deep and potentially unmanageable tension’ between profit-seeking investors and directors tasked with protecting humanity from the risks of advanced AI.
The paper notes that critics have traditionally worried these governance models would eventually drift toward commercial priorities as companies seek capital, partnerships and talent.
However, the authors argue the opposite risk has received far less attention: that insulated directors may overreach, damaging both investors and the mission they were appointed to protect.
The paper labels this possibility ‘Ben & Jerry’s risk’.
The term refers to the breakdown of governance arrangements at Ben & Jerry’s following Unilever’s acquisition of the company in 2000.
As part of that transaction, Unilever agreed to establish an independent board tasked with safeguarding Ben & Jerry’s social mission and brand integrity. The board was largely self-perpetuating and insulated from direct investor control.
For roughly two decades, tensions between the board and Unilever remained largely private. That changed in 2021 when Ben & Jerry’s announced it would stop selling products in Israeli-controlled territories.
The decision triggered political backlash, litigation, divestments by several US state pension funds and a broader conflict between the independent board and Unilever.
According to the paper, the fallout contributed to billions of dollars in lost market value at Unilever and ultimately failed to achieve the board’s stated objective.
Unilever later transferred rights enabling continued sales of Ben & Jerry’s products in Israel and the territories, effectively bypassing the board’s efforts.
The authors describe this outcome as ‘double trouble’: harming investors while simultaneously undermining the mission itself.
OpenAI’s near-crisis
The paper argues that a similar pattern emerged at OpenAI during the attempted removal of chief executive Sam Altman in November 2023.
OpenAI’s nonprofit-controlled structure was originally designed to ensure the company’s AI development remained aligned with humanity’s interests rather than investor returns.
However, Fried and Reiter argue that the board’s attempt to remove Altman exposed the instability that can emerge when directors are largely insulated from investor accountability.
The attempted leadership change threatened OpenAI’s commercial partnerships, financing arrangements and employee retention. The crisis also risked substantial losses for investors.
Ultimately, Altman returned as chief executive, most of the original board departed and several prominent AI safety researchers later left the company.
The authors argue this episode mirrored the Ben & Jerry’s experience because the governance intervention not only endangered investor value but also weakened the safety-focused mission the directors were trying to protect.
The paper notes that OpenAI’s governance structure is particularly difficult to unwind because of its nonprofit origins and related regulatory obligations.
Anthropic’s ‘kill switch’
In contrast, Anthropic’s governance framework includes what the paper describes as a ‘kill switch’.
Under Anthropic’s structure, investors reportedly retain the ability to remove mission-focused guardians through a supermajority mechanism.
The authors argue this arrangement reduces the risk of severe investor harm by limiting the power of insulated directors.
At the same time, they suggest it may weaken the effectiveness of those directors in pursuing AI safety objectives because investors ultimately retain a path to override them.
For governance practitioners, the comparison highlights a broader tension increasingly relevant across mission-driven companies: how to preserve long-term societal objectives without creating boards that lack meaningful accountability.
Governance implications
The paper arrives as governance structures at AI companies face growing scrutiny from investors, regulators and policymakers.
Many of the sector’s leading firms are pursuing corporate forms that diverge sharply from traditional shareholder-centric governance models.
Supporters argue these structures are necessary because advanced AI could create systemic societal risks that conventional governance frameworks are ill-equipped to manage.
But Fried and Reiter caution that governance systems relying on insulated mission guardians may introduce new risks rather than resolving existing ones.
Their analysis suggests boards operating without clear accountability mechanisms can become destabilizing forces inside companies, particularly when missions are broad, politically contentious or difficult to measure.
For boards and investors, the paper reinforces the importance of clearly defining governance authority, escalation processes and director accountability when balancing commercial performance with wider stakeholder or societal objectives.
It also raises practical questions about whether highly mission-driven governance models remain workable once companies scale, attract outside capital and become systemically important.
As AI companies continue to raise billions of dollars while positioning themselves as guardians of public interest, the debate over who ultimately controls these firms – investors, founders or mission-driven directors – is likely to intensify.
The paper suggests that tension may not be temporary but embedded directly into the governance architecture itself.
Natalie Bannerman
Natalie is a former telecoms and infrastructure journalist, a role she held for nearly seven years. Before this, she worked in the B2C startup space, covering lifestyle, arts and culture reporting. As senior reporter for Governance Intelligence she...
