UPDATE

Not For Private Gain

Nov 2025

Download PDF

Introduction

Comparison of the outcome to OpenAI’s original proposal

How This Might Go Well

Conclusion

Signatories

Footnotes

Statement on OpenAI’s Restructuring

Nearly a year after its original proposal, OpenAI has completed its corporate restructuring. Before agreeing not to object to the restructuring, the Attorneys General of California and Delaware extracted 20 concessions from OpenAI, enumerated in memoranda of understanding.¹ In light of those concessions and OpenAI’s corporate filings, we find that the outcome—while still inherently contrary to OpenAI’s charitable mission—is a substantial improvement over OpenAI’s original restructuring proposal. Most importantly, the directors of the new Public Benefit Corporation (PBC) are required to prioritize the charitable mission over shareholders’ interests when making safety and security decisions; OpenAI Foundation (“the nonprofit”) has approval rights over safety and security decisions, including the power to halt the release of a model; and the nonprofit board has the sole right to hire and the non-exclusive right to fire PBC directors.

The first section of this document compares the restructuring as consummated to OpenAI’s original proposal and provides our assessment of the outcome.² Potentially more important than the legal outcome, however, is how OpenAI implements it. The second section describes five indicators to track over the next year that would signal whether OpenAI is earnestly viewing its new structure as a vehicle to advance its charitable mission. If these indicators are not met, public advocates will take notice.

We are heartened to hear that public advocates are not alone in this regard. As California Attorney General Rob Bonta declared, “We will be keeping a close eye on OpenAI to ensure ongoing adherence to its charitable mission.”³ And as Delaware Attorney General Kathleen Jennings reminded, “Anyone who is familiar with our work knows we are not shy to go into the courtroom to benefit the public if we need to.” If OpenAI fails to meet the letter or spirit of its agreements with the Attorneys General, we expect the Attorneys General will also take notice.⁴

I. Comparison of the outcome to OpenAI’s original proposal

OpenAI’s original restructuring proposal envisioned a typical PBC and an entirely separate nonprofit focused on philanthropic initiatives with no explicit connection to AI safety or security. As the December 2024 announcement states: “The PBC will run and control OpenAI’s operations and business, while the non-profit will hire a leadership team and staff to pursue charitable initiatives in sectors such as health care, education, and science.”⁵ The restructuring, as implemented nearly a year later, contains substantial improvements, addressing many—though far from all—of the concerns we raised in our May 12 letter.⁶ Importantly, most of these improvements are requirements of the Attorneys General, meaning OpenAI cannot disregard or dilute them without risking legal action by the Attorneys General. This section highlights eight components of OpenAI’s restructuring, comparing the outcome with OpenAI’s original restructuring proposal.⁷

Component of Structure

Previous structure (2019-2025)

OpenAI’s Original Restructuring Proposal
(Dec. 2024)

Outcome
(Oct. 28, 2025)

1. Profit motives are subordinate to charitable mission

Yes

Yes for safety and security decisions; otherwise no

2. Nonprofit control

Yes, management-level

Yes for safety and security decisions, and sole power to hire / non-exclusive power to fire PBC directors; otherwise no

3. Attorney General oversight

Yes, since nonprofit has management-level control

Yes over nonprofit’s control functions (see above) and compliance with their 20 requirements of the restructuring

4. Majority-independent board commitment

Yes

No requirement

Yes

5. Nonprofit governance of AGI when developed

Yes

6. Stop-and-assist commitment from Charter

Yes

No by default

Yes, along with rest of the OpenAI Charter

7. Nonprofit’s financial stake

~2% equity (from 2023 reporting); right to all above-cap profits

~9-19% of valuation (based on rumored $30 billion); no above-cap profits

26% of valuation ($130 billion); no above-cap profits but warrant of unspecified value

8. Philanthropy

Little historically; future plans unknown

Yes, stated focus on health care, education, and science

Yes, stated focus on health and disease, and technical solutions to AI resilience

Table 1. Comparing components of OpenAI’s previous structure with its original restructuring proposal (Dec. 2024) and completed restructuring (Oct. 28, 2025).

1. Profit motives are subordinate to charitable mission

OpenAI’s original proposal: The charitable mission would not take precedence over OpenAI’s commercial goals, in contrast to its previous structure. At best, the PBC’s directors would be required to “balance” the mission against shareholders’ interests.⁸ But even that assumes the charitable mission would have been adopted as the PBC’s public benefit goal, to which OpenAI had not committed. Indeed, it is unclear if even the nonprofit would retain the same mission in light of OpenAI’s description of its focus as “charitable initiatives in sectors such as health care, education, and science.”⁹

Outcome: As required by the Attorneys General,¹⁰ the PBC’s Certificate of Incorporation explicitly enshrines the charitable mission as the PBC’s public benefit goal, and the PBC’s directors must solely consider the charitable mission with respect to safety and security issues.¹¹ The Attorneys General define “safety and security issues” to include “all actions and decisions of the members of the Safety and Security Committee,” as discussed below.¹²

Our assessment: This is a critical foundational requirement that informs the scope of items #2 (nonprofit control) and #3 (Attorney General oversight). We are pleased that the Attorneys General recognize the inherent conflicts between the interests of shareholders and the public (who are the nonprofit’s beneficiaries) regarding safety and security decisions. We nonetheless regret that the mission-primacy requirement does not explicitly extend to other decision categories where commercial interests might conflict with the mission, such as lobbying.

2. Nonprofit control

OpenAI’s original proposal: The nonprofit would have no oversight role or control over the PBC beyond the powers of a minority shareholder. As the December 2024 announcement states: “The PBC will run and control OpenAI’s operations and business, while the non-profit will hire a leadership team and staff to pursue charitable initiatives in sectors such as health care, education, and science.”¹³

Outcome: As required by the Attorneys General, the nonprofit will have the sole power to hire and the non-exclusive power to fire the PBC’s directors, powers they are required to exercise in a manner consistent with their fiduciary duties to the mission and its public beneficiaries.¹⁴ In addition to the nonprofit, PBC directors can also be fired by a two-thirds vote of the PBC’s shareholders, with the caveat that the other shareholders cannot remove more than one-third of the PBC’s directors in any twelve-month period.¹⁵

In addition, the nonprofit will have approval rights over safety and security decisions through its Safety and Security Committee (SSC).¹⁶ The SSC was a committee of the former nonprofit board,¹⁷ which must be retained under the new nonprofit board. The SSC’s powers are not fully defined in the agreement, but must be “consistent with current practice prior to the Recapitalization” and include the following:

overseeing and reviewing the safety and security processes and practices of the Corporation and its controlled affiliates with respect to model development and deployment [and] . . . the authority to require mitigation measures—up to and including halting the release of models or AI systems even, for the avoidance of doubt, where the applicable risk thresholds would otherwise permit release.¹⁸

The scope of the SSC’s powers in OpenAI’s previous structure are apparently described in a non-public document from September 2024.¹⁹

The SSC must be chaired by a nonprofit director who is not also serving as a PBC director.²⁰ The nonprofit must have access to information it requires to advance the charitable mission, including the PBC’s advanced research, intellectual property, AI models, and employees.²¹

The nonprofit’s oversight powers are protected from modification and dilution via robust pre-approval rights.²²

Our assessment: While a significant downgrade from the nonprofit’s management-level control of OpenAI’s previous commercial entity, these are meaningful improvements from the original proposal. Granting the nonprofit approval rights over safety and security decisions is the most important change from OpenAI’s original proposal. While the scope of these powers appears to be broad, they are nonetheless either under-defined or further defined in nonpublic documents. The mission will be best served if these powers are broad in scope and protected from dilution.

Whether the nonprofit will effectively exercise its approval rights over safety and security decisions is unclear, however. As discussed in the following section, the nonprofit’s SSC will need to be well-resourced to be effective.

Although the nonprofit’s more-general power to hire and fire the PBC’s directors is significant, it might not be as powerful as it initially appears for two reasons. The first reason relates to the directors’ dual loyalties. The power to hire and fire PBC directors would be powerful in the hands of directors loyal solely to the charitable mission. Unfortunately, with one exception now (and as few as two exceptions a year from now²³), the two boards comprise the same people. It will be essential that the directors are vigilant about the capacity in which they are acting and the different duties and beneficiaries entailed, and we are concerned that doing this consistently and effectively would be difficult for even the most informed and conscientious directors acting entirely in good faith. The Attorneys General have required that the directors with dual roles attend annual trainings about their fiduciary duties,²⁴ which we hope will focus in significant part on best practices for managing dual loyalties. If the directors on both boards end up being loyal primarily to the PBC, the nonprofit’s power to hire and fire PBC directors—the basis of its nominal control of the PBC—would be effectively meaningless.²⁵

Second, giving other shareholders (with a two-thirds majority) the power to fire directors means they can remove directors they perceive to be too focused on the charitable mission, relative to the shareholders’ own interests. Without the limit to the shareholders’ power—that they cannot fire more than one-third of the board in a twelve-month period—shareholders would theoretically be able to negate the nonprofit’s hiring power by immediately firing any director of whom they do not approve. While this limit (and the nonprofit’s control of a significant percentage of common shares) prevents gross shareholder abuse, any shareholder firing power weakens the nonprofit’s control of the PBC.²⁶

3. Attorney General oversight

OpenAI’s original proposal: Because in this proposal the nonprofit would have no control over the PBC, the Attorneys General would likewise have no power to oversee the PBC through their power over charities. In the event the PBC were to grossly disregard the charitable mission, only shareholders would have recourse.²⁷

Outcome: The nonprofit will have the power to hire and fire the PBC’s directors and approval rights over safety and security issues, including “the authority to require mitigation measures—up to and including halting the release of models or AI systems.”²⁸ The Attorneys General have the power to ensure the nonprofit directors are exercising those responsibilities in a manner consistent with their fiduciary duties to the mission and its beneficiaries. In principle, this means the Attorneys General can intervene if the nonprofit directors appoint or fail to remove PBC directors who undermine OpenAI’s mission, or if the SSC fails to require adequate mitigation measures or halt dangerous model releases for commercial reasons.

The Attorneys General are also entitled to monitor OpenAI’s compliance with the 20 requirements in their memoranda of understanding. To support their compliance monitoring, members of the nonprofit board’s Mission & Strategy Committee must be available to meet with the offices of the Attorneys General semiannually, and senior PBC executives must be available to meet with their offices at least quarterly. The purpose of these meetings is to provide the Attorneys General with “timely information about [OpenAI’s] operations and progress towards its mission.”²⁹ OpenAI must provide the Attorneys General 21-day advance notice of any transaction that narrows the nonprofit’s rights vis-a-vis the PBC, including any change-of-control transaction and any change to the PBC’s public benefit mission.³⁰ The Attorneys General may employ experts and consultants, at OpenAI’s expense, to help evaluate these noticed transactions.³¹

Our assessment: This is a significant improvement over OpenAI’s original proposal, which would have eliminated nearly all Attorney General oversight of the PBC’s operations. How it compares to the previous structure is more complicated. In one respect, the Attorneys General have a much narrower oversight role, as they are the primary regulators of nonprofits (not PBCs), and the nonprofit now has less control over OpenAI’s commercial operations.

On the other hand, the outcome might mean more Attorney General oversight in practice for two reasons. First, OpenAI is now bound to the terms of their agreements with the Attorneys General. Going forward, Attorney General oversight will focus not only on general violations such as breaches of fiduciary duty, but also violations of the specific terms of the agreements, such as whether OpenAI has met its public-reporting requirements.³² Second, whereas active oversight was previously a power the Attorneys General could exercise, going forward it is one they will exercise through their reporting and semiannual meeting requirements.

4. Majority-independent board requirement

OpenAI’s original proposal: Delaware corporate law does not require that the board of a PBC be majority independent, and OpenAI’s proposal did not mention it.

Outcome: As required by the Attorneys General: “The PBC Board will be composed of a majority of independent directors, reinforcing oversight.”³³ Specifically, the independent directors “will not be employees or members of management, and, in the determination of the PBC Board, will have no relationship or interest that could compromise their judgment—ensuring strong, objective oversight that reinforces accountability and mission alignment.”³⁴

Our assessment: Although good in principle, we are concerned that some members of the existing board are considered “independent” by OpenAI³⁵ despite having relationships with OpenAI through other entities that could bias their judgment.³⁶ We are also concerned that whether a director is independent will be “in the determination of the PBC Board.” As discussed in the following section, we hope the Attorneys General will scrutinize the PBC board’s decisions regarding director independence.

5. Nonprofit governance of AGI when developed

OpenAI’s original proposal: The proposal did not mention nonprofit governance of AGI, despite OpenAI’s previous commitment that “AGI technologies are explicitly reserved for the Nonprofit to govern.”³⁷ As a minority shareholder of the PBC, the nonprofit would not have the right to govern AGI by default.

Outcome: The nonprofit has no right to govern AGI beyond its approval rights over safety and security decisions and the power to hire and fire PBC directors. Under the previous structure, Microsoft’s rights to OpenAI’s intellectual property excluded AGI technologies, leaving open the possibility OpenAI would not commercialize AGI if that best served the interests of humanity. That is no longer the case.³⁸

Our assessment: Although disappointed, we recognize that nonprofit governance of AGI might be difficult to reconcile with OpenAI’s IPO ambitions, and therefore it would have been a significant concession. Nonetheless, it is a poignant loss. OpenAI was founded as a nonprofit to ensure AGI would benefit all of humanity; nonprofit governance of AGI was a core component of its end goal. It launched its commercial subsidiary in 2019 only as a means to finance this very end goal. We are saddened to see OpenAI’s commercial goals elevated from means to end.

6. Stop-and-assist commitment from Charter

OpenAI’s original proposal: The proposal did not mention either re-adopting the OpenAI Charter as the “principles [the PBC would] use to execute on OpenAI’s mission” or the specific commitment contained in the Charter to stop competing with and start assisting “a value-aligned, safety-conscious project [that] comes close to building AGI before [OpenAI does].”³⁹

Outcome: The Attorneys General have required that the PBC adopt the OpenAI Charter as “the principles the PBC will use to execute [its] Mission.”⁴⁰ According to an attorney in the office of the Delaware Attorney General: “The charter was important to us [and was] one of the key concessions that we got.”⁴¹

Our assessment: We agree that this is an important concession for at least two reasons. First, by adopting the Charter, OpenAI has re-committed to stop competing with and start assisting another frontier AI project under certain circumstances, a commitment made explicitly because OpenAI’s leadership were “concerned about late-stage AGI development becoming a competitive race without time for adequate safety precautions.”⁴² To legally bind the PBC, however, this commitment would need to be incorporated into the PBC’s Certificate of Incorporation, which is not currently the case.

Second, OpenAI’s mission as articulated in the Charter hews closer to its original meaning, which has become increasingly lost as OpenAI aggressively pursues commercial endeavors. For example, one might read OpenAI’s description of its mission in a recent policy submission to mean that making useful AI tools is the primary means by which it will advance its mission.⁴³ Such a reading would not meaningfully distinguish OpenAI, with its unique nonprofit origins, from conventional for-profit AI companies. Recognizing that the Charter contains the principles the PBC will use to execute the mission protects the spirit of the mission from further distortion or dilution. For example, the stop-and-assist commitment in the Charter highlights the difference between building AGI and ensuring AGI is built safely and for the benefit of humanity. It is now unambiguous that OpenAI’s mission is the latter. Not only might not competing be compatible with OpenAI’s mission, under certain circumstances, the mission might require it.

The Charter’s principles include: broadly distributed benefits; long-term safety; technical leadership; and cooperative orientation.⁴⁴ These principles, as expounded upon in the Charter, should serve as a north star for the nonprofit directors and inform how the Attorneys General assess OpenAI’s compliance with the terms of their agreement.

7. Nonprofit’s financial stake

OpenAI’s original proposal: The proposal did not mention the profit caps,⁴⁵ which OpenAI had previously heralded as an important part of its charitable mission by ensuring that any outsized profits from AGI are used to benefit humanity.⁴⁶ OpenAI’s May 5 announcement, however, made it clear the profit caps would be removed.⁴⁷ While OpenAI did not announce a value of the nonprofit’s proposed stake, it was reported to be approximately $30 billion in January 2025.⁴⁸ At that time, OpenAI had a valuation of $157 billion from October 2024 and was raising additional funds at a reported $340 billion valuation,⁴⁹ putting the $30 billion roughly in the 9-19% of overall equity range.

Outcome: The nonprofit’s equity stake is valued at approximately $130 billion, or 26% of overall common equity.⁵⁰ The nonprofit is also receiving a warrant that will grant it an undisclosed number of additional shares in the PBC (described by OpenAI as “significant additional equity”) if the value of the PBC’s shares increases by at least 10x over the next 15 years.⁵¹ According to OpenAI, “[w]ith its equity stake and the warrant, the [nonprofit] is positioned to be the single largest long-term beneficiary of OpenAI’s success.”⁵²

Our assessment: The 26% ($130 billion) stake is of course larger than the 9-19% ($30 billion) stake reportedly under discussion in January. Our overall take, however, is largely dependent on the economics of the warrant, which have not been fully disclosed. OpenAI also has not disclosed the details of the current profit caps, without which a full assessment of the nonprofit’s stake is not possible. In the unlikely event the combined 26% stake and warrant approximates the expected value of the nonprofit’s stake in OpenAI’s previous commercial entity⁵³ and its right to above-cap profits—based on the financial projections OpenAI has shared with investors—we would be pleased with this outcome. Otherwise, it is a loss for the public, for whose benefit these funds were promised.

We are also concerned that the trigger for the warrant—10x value increase within 15 years—creates an incentive for the nonprofit to favor speed in situations where the public’s interests might be best served by caution. Such an incentive might bias the SSC’s determinations regarding model deployment.

8. Philanthropy

OpenAI’s original proposal: The nonprofit would “hire a leadership team and staff to pursue charitable initiatives in sectors such as health care, education, and science.”⁵⁴ OpenAI did not comment on the speed or scale at which these initiatives would proceed.

Outcome: In addition to the previously announced $50 million commitment “to support nonprofits and mission-focused organizations working at the intersection of innovation and public good,”⁵⁵ OpenAI has announced a $25 billion commitment across two areas: health and curing diseases, and technical solutions to AI resilience.⁵⁶ It has not commented on when these funds would be disbursed.

Our assessment: Executed well, OpenAI’s philanthropic plans are laudable, and we are pleased to see the more-explicit focus on AI-safety-related problems. Public advocates, however, have expressed concern that OpenAI’s philanthropic plans are self-serving.⁵⁷ Managing conflicts of interest with the PBC will likely be a challenge for all of the nonprofit’s philanthropic work, but it will be particularly challenging for the nonprofit’s AI resilience program. For example, OpenAI might be biased against valuable technical AI safety work that could identify risks in OpenAI’s models, creating commercial costs; and it might be biased in favor of training programs that address unemployment risks by teaching people to use ChatGPT, generating commercial benefits. It is incumbent upon OpenAI to prove its critics wrong by establishing that it can navigate these conflicts of interest.

II. Looking forward: Indicators to track

How OpenAI’s new structure will work in practice depends on how OpenAI implements it. Below are five indicators of whether OpenAI is taking the steps necessary to advance its charitable mission within its new structure.

Directors on both boards take their dual loyalties seriously. With one exception, the current directors now have two roles, as director of the nonprofit and of the PBC. Each role has different duties and beneficiaries. To have nonprofit control of OpenAI in any meaningful sense, it is essential that the directors always know in which capacity they are acting and how that impacts their actions. We encourage OpenAI to include in their required public reporting information about how the directors are managing their dual loyalties. It might ultimately prove too difficult for directors to separate their loyalties. Among the most consequential tests of the nonprofit board’s loyalty is whether it can recognize that problem and take the steps necessary to resolve it.

Appoint a new, independent nonprofit director with relevant expertise. The nonprofit’s responsibilities will be different in the new structure, and the current directors were not selected with these responsibilities in mind. While the nonprofit was previously responsible for managing the entire business, its oversight responsibilities going forward are more focused on safety and security decisions, as those are the decisions it has the power to veto. The Attorneys General have required that within one year, the nonprofit has a second director who does not also serve on the PBC board.⁵⁸ Appointing a new director with expertise relevant to its new responsibilities would provide evidence that the nonprofit is taking those responsibilities seriously.

Appoint an independent and qualified nonprofit CEO. OpenAI has not commented on who will be the CEO of the nonprofit. Sam Altman, as the presumptive CEO of the PBC, should be ineligible. Appointing a CEO that is both independent of the PBC and has deep expertise relevant to the nonprofit’s responsibility to oversee the PBC’s safety and security decisions would evidence that the nonprofit directors appreciate the importance of their responsibilities.

Resource the Safety and Security Committee. Through its Safety and Security Committee, the nonprofit has significant power and responsibilities over the PBC’s safety and security decisions. It is unrealistic, however, to expect part-time directors alone to effectively exercise that power over a company with thousands of employees working at the frontier of a rapidly changing technology. Allocating significant resources—including full-time staff and consulting experts—to the SSC would signal that the nonprofit directors appreciate the magnitude and scope of the SSC’s responsibilities and intend for it to be more than a rubber stamp.

Improve public transparency. OpenAI has not historically provided its nonprofit beneficiaries—the public—the information necessary to understand how OpenAI’s decisions are impacting their interests.⁵⁹ The Attorneys General have required that the PBC make at least annual public disclosures of its progress toward the charitable mission.⁶⁰ The nature of those disclosures, among others, will evidence whether OpenAI views this as a box-checking exercise or a genuine opportunity to inform its beneficiaries. For example, OpenAI should immediately disclose (i) information about the economics of the nonprofit’s warrant; (ii) any nonpublic documents describing the scope of the SSC’s responsibilities, including the Unanimous Written Consent, dated September 15, 2024;⁶¹ and (iii) any information regarding how it would operationalize the Charter’s stop-and-assist commitment.

Finally, robust oversight by the Attorneys General would substantially improve our expectations for the restructured OpenAI, ensuring OpenAI is following both the letter and spirit of their agreements. The Attorneys General should not rely only on OpenAI’s representations of compliance. For example, we are concerned that the PBC board might adopt an unduly narrow reading of the “no relationship or interest that could compromise their judgment” standard for director independence. For important safety and security decisions, such as model deployment, we hope the Attorneys General require contemporaneous documentation of the SSC’s determinations, evidencing it has solely and effectively considered the charitable mission.

III. Conclusion

OpenAI’s restructuring plainly does not advance its charitable mission. Nonetheless, the final version is a substantial improvement upon the original proposal. In most respects, OpenAI has retained the ability to ensure AGI is built safely and for the benefit of humanity, while losing some of the guardrails preventing it from deviating from that mission. By proposing the restructuring, OpenAI leadership implicitly claimed that these guardrails are unnecessary and that OpenAI will advance the charitable mission even more effectively without them. We hope their actions going forward show our concerns were unnecessary.

We are deeply grateful for everyone whose efforts improved this outcome. Thank you to everyone who signed our April open letter; to the EyesOnOpenAI coalition, former OpenAI employees, and all the other public advocates who stood up and made their voices heard; and to Attorneys General Bonta and Jennings, and the staff at each of their offices, for their hard work to protect the public’s interests in the face of tremendous pressure.

Respectfully submitted,

Page Hedley

OpenAI 2017-2018

Page Hedley

OpenAI 2017-2018

Page Hedley

OpenAI 2017-2018

Tyler Whitmer