OPENAI FOUNDATION

Institutional Archetype

THE OWNER OF THE REFEREE’S WHISTLE — The OpenAI Foundation is the nonprofit that remained in control of OpenAI when the company became a public benefit corporation. It appoints the company’s directors, holds roughly a quarter of its stock, and houses the board committee that, on the Attorneys General’s record, can halt a model release. It is also a grantmaker in the evaluation and biosecurity field, which means the owner of a lab now pays some of the people who test labs. The file is open because the same board appears on every side of that sentence, and because its own announcements, the two state attorneys general and one of its grantees have all written the arrangement down.

Mandate & Origin

The Foundation describes itself in one sentence: “The OpenAI Foundation is a charitable organization dedicated to ensuring AI benefits all of humanity” (OpenAI Foundation, 9 Sep 2026). OpenAI’s structure page gives the lineage: “OpenAI was founded in 2015 as a nonprofit,” a for-profit subsidiary created in 2019 “has always been governed and controlled by the nonprofit,” and under the structure announced on 28 October 2025 “The nonprofit is now the OpenAI Foundation” while the for-profit became OpenAI Group PBC (OpenAI, Our structure). Chair Bret Taylor’s announcement of the same day: “The OpenAI Foundation controls the for-profit business, which was established in 2019” (Bret Taylor, 28 Oct 2025).

The recapitalization went through two state reviews. Delaware’s Attorney General opened hers on 9 October 2024, retained independent counsel and “an independent financial advisor, Moelis & Company LLC,” and coordinated with California (Delaware Department of Justice, 28 Oct 2025). OpenAI says the transaction “was completed after nearly a year of engaging in constructive dialogue with the offices of the Attorneys General of California and Delaware” (OpenAI, Our structure). The Foundation’s current website carries three programs, Life Sciences and Curing Diseases, AI Resilience, and Civil Society and Philanthropy, and a Jobs and Economic Futures program announced in May 2026 (OpenAI Foundation; OpenAI Foundation, 27 May 2026).

Funding & Backers

An endowment in the company’s stock. “As of the closing of the recapitalization, the OpenAI Foundation holds a 26% equity stake in OpenAI Group, worth approximately $130B based on OpenAI Group’s current valuation,” plus “a warrant that allows it to receive additional shares in OpenAI Group if it hits a valuation milestone.” Microsoft “holds roughly 27% of OpenAI Group, and the remaining 47% is held by current and former employees and investors” (OpenAI, Our structure). The same page says the Foundation board “were advised by independent financial advisors, and the terms of the recapitalization were unanimously approved by the board.” Taylor put the consequence plainly: “The more OpenAI succeeds as a company, the more the non-profit’s equity stake will be worth, which the non-profit will use to fund its philanthropic work” (Bret Taylor, 28 Oct 2025).

What it has said it will spend. “The OpenAI Foundation will initially focus on a $25B commitment across two areas”: health and curing diseases, and “Technical solutions to AI resilience” (Bret Taylor, 28 Oct 2025). On 24 March 2026 the Foundation wrote that it “expects to invest at least $1 billion across life sciences and curing diseases, jobs and economic impact, AI resilience, and community programs” over the next year (OpenAI Foundation, 24 Mar 2026).

What it has announced paying.

The Foundation’s own pages do not list the AI Resilience grantees by name; the one resilience-field recipient the record shows is the grantee’s own announcement, which is the SecureBio row. The AI-safety evaluator field is otherwise funded elsewhere: see Open Philanthropy (Coefficient Giving), Schmidt Sciences and the Frontier Model Forum for the funders whose grants are on the record.

Institutional Voice & Intent

The Foundation writes essays, not press releases, and its register is the long program post signed by the program lead.

• Chair Taylor, 24 March 2026, on safety work: “We want AI systems to be safer by default. That means supporting independent testing and evaluations, developing new and stronger industry standards, and funding foundational research that helps to avoid safety issues or to detect and address them early” (OpenAI Foundation).
• The AI Resilience post, 1 June 2026: “The early OpenAI team believed that ensuring AI benefits society depended primarily on solving the technical alignment problem. That remains critical—and central to our work—but we now believe it is only one part of the puzzle” (OpenAI Foundation).
• The same post, on who should evaluate: “AI companies are investing substantial resources in model safety. However, the importance of this challenge calls for a broader, more robust ecosystem: independent institutions to evaluate model safety, public infrastructure to verify models’ safe deployment in practice, and continued advances in alignment science that advance the field broadly” (OpenAI Foundation).
• And its analogy: electricity needed “independent testing bodies like Underwriters Laboratories,” and “No general purpose technology ever made itself safe” (OpenAI Foundation).

Stated intent: Use the endowment to fund science, civil society and an AI-resilience ecosystem, including independent evaluators, so that safety does not depend on any one company’s safeguards.

Observed intent: Seat the governance of the lab’s safety decisions in the nonprofit, per the Attorneys General’s conditions, and fund the biosecurity and evaluation field with proceeds of a stake whose value rises with the lab’s own success. The Foundation says the second half itself.

Gap: The record shows the structure and the grants and cannot show what either does to anyone’s conclusions. SecureBio’s Detection Director, the one grantee who has written the question out, states it as an open risk and then states the controls: “While it looks to me like OAIF is making its own decisions, the two organizations are closely linked in a way that goes beyond the shared ‘OpenAI’ name: the OpenAI Foundation’s endowment is a ~1/4 stake in OpenAI PBC, and they have almost identical boards” (Jeff Kaufman, SecureBio Detection Director). The Foundation’s counterpart statement is that the grant is restricted to Detection and “places no constraints on what any part of SecureBio can research, publish, or say” (SecureBio). The gap is structural: a funder, a committee and a stockholder that share one board and one stock price, with the insulation written into two state memoranda and one grant letter.

Interest: The Foundation does not describe its positions as self-interest. What each would pay it, or the company whose stock it holds, is set out from its own pages and the filings.

• Independent evaluation and testing as the safety layer. Position, the Foundation’s: “independent institutions to evaluate model safety, public infrastructure to verify models’ safe deployment in practice.” What it pays: the committee the Foundation houses and the company it owns both operate under the Attorneys General’s conditions; an outside evaluation layer is the thing a regulator, a customer or a court can point to besides the company’s own word, and the Foundation’s stake gains from the company’s continued deployment (“The more OpenAI succeeds as a company, the more the non-profit’s equity stake will be worth”). Best case, the Foundation’s: “No general purpose technology ever made itself safe,” with Underwriters Laboratories as the precedent. Convergence: the same layer is funded by Open Philanthropy, Schmidt Sciences and the Frontier Model Forum’s members; Jacob Trefethen moved from Coefficient Giving to lead the Foundation’s science program.
• Pathogen-agnostic biosecurity. Position: “we will prioritize pathogen-agnostic biosecurity solutions.” What it pays: the $17.2M to SecureBio’s Detection arm is the documented instance; the company’s GPT-6 Astra assessment was run by the grantee’s other arm. The grantee’s own statement of the risk the position creates: “the real worry is a subtle drift towards being more generous without anyone explicitly asking for anything,” and its Detection Director’s worry that stronger defenses let developers “push ahead faster.” Best case, the grantee’s: “The world will soon need robust defenses in place that don’t depend on any single company’s safeguards.” Convergence: the grantee’s primary funders are Coefficient Giving, the Survival and Flourishing Fund and Sentinel Bio (SecureBio).

Interest is a documented fact about each position, not a claim about motive, and not a verdict that the argument is wrong. Each best case is stated at the strength its author states it.

Position in the Apparatus

• The committee that can say no. The Delaware Attorney General’s statement of 28 October 2025: the Foundation’s “board-level Safety and Security Committee … will remain a committee of the NFP and not be moved to the PBC,” with “the power and authority to require mitigation measures—up to and including halting the release of models or AI systems—even where the applicable risk thresholds would otherwise permit release.” Its chair is Zico Kolter: “Initially, this will be the current committee chair, Mr. Zico Kolter” (Delaware Department of Justice). OpenAI’s structure page: “The chair of the SSC—Dr. Kolter—will serve exclusively on the OpenAI Foundation board” (OpenAI, Our structure). Kolter co-founded Gray Swan, which tests OpenAI’s models.
• The September 2026 appointment. “We’re announcing the appointment of Paul Christiano to the OpenAI Foundation Board. He will be a non-voting observer on the OpenAI Group PBC Board,” and he “will also join the Safety and Security Committee (SSC) of the Foundation Board, working alongside its chair, Zico Kolter.” The post names his seat at NIST’s Center for AI Standards and Innovation, where he is “a Senior Tech Advisor,” and adds: “Paul will recuse himself from all OpenAI-related matters as well as all model evaluations” (OpenAI Foundation, 9 Sep 2026). See Paul Christiano, US CAISI and the Alignment Research Center, which he founded and whose evaluations team became METR.
• The company it controls. “Through special voting and governance rights held solely by the OpenAI Foundation, the OpenAI Foundation appoints all members of the board of directors of OpenAI Group and can replace directors at any time” (OpenAI, Our structure). See OpenAI and its chief executive, Sam Altman, who sits on the Foundation board.
• A seat for a former intelligence chief. The board list names “Retired U.S. Army General Paul M. Nakasone” (OpenAI, Our structure); see Paul Nakasone.
• Staff from the evaluator-funding world. The 24 March 2026 post names Jacob Trefethen as Head of Life Sciences and Curing Diseases, “joining from Coefficient Giving, where he oversaw more than $500 million in grantmaking to science and health”; Wojciech Zaremba, “a co-founder of OpenAI,” as Head of AI Resilience; and Anna Makanju as “Head of AI for Civil Society and Philanthropy.” It also says “The Foundation Board is also searching for an Executive Director” (OpenAI Foundation).
• A bank chief on the board, a bank advisory role on the committee chair. The Foundation’s July 2026 post names “David Vélez and Robin Vince to the boards of the OpenAI Foundation and OpenAI Group PBC,” and calls Vince “Chairman and CEO of BNY” (OpenAI Foundation, 21 Jul 2026). Kolter’s own biography lists “an advisor to BNY” (Kolter, Bio).

Actions & Leadership Choices

CONDUCT: STRUCTURAL — OWNS THE LAB, SEATS ITS SAFETY COMMITTEE, FUNDS THE TESTERS. No wrongdoing is alleged. The structural point is that one nonprofit board is the controlling stockholder of a frontier lab, the home of the committee that can halt its releases, and a funder in the biosecurity and evaluation field, and that its own announcement of a safety-committee appointment says the committee “provides governance over safety and security practices across all of OpenAI, including OpenAI Group PBC.” The strongest case for the arrangement is the documentary one: the powers are written into a Delaware statement and a California memorandum, not left to goodwill.

What the two states recorded. Delaware’s conditions include that “PBC directors will be required to consider only the mission (and may not consider the pecuniary interests of stockholders or any other interest) with respect to safety and security issues related to the OpenAI enterprise and its technology”; “Within one year of the recapitalization, the NFP Board will have at least two directors (including the Chair of the Safety and Security Committee) who will not serve on the PBC Board”; and “The Attorney General will be provided with advance notice of significant changes in corporate governance” (Delaware Department of Justice). California’s memorandum, signed by both parties on 27 October 2025, adds that “The PBC and NFP will enter a contractual agreement providing the SSC an effective approval right”; that the company “will be contractually obligated to provide the NFP with those reasonable services and resources without compensation”; that “The decision whether to relinquish the Class N Common Stock, if ever, is in the control of the NFP Board”; and that the NFP “will provide at least 21 days’ prior written notice to the Attorney General before consenting to: (a) a change of control of the PBC” (California Department of Justice, MOU, 27 Oct 2025).

The grantee that wrote down the problem. SecureBio announced the grant on 21 August 2026: the Foundation “has granted SecureBio Detection $17.2M to reduce our end-to-end time from fourteen to three days.” Its AI team’s conflict disclosure in the GPT-6 Astra assessment reads: “The OpenAI Foundation, a 501(c)(3) nonprofit that holds equity in OpenAI, granted SecureBio’s Detection arm $17.2M to build an early-warning pathogen detection system” (SecureBio; SecureBio, GPT-6 Astra assessment). The Detection Director adds that the company’s own payment for an evaluation is the tighter link: “OpenAI PBC covered SecureBio’s costs for the GPT 5.5 evaluation above” (Kaufman). The cross-evaluator picture is on The Evaluator Alliances; the grantee’s own file is SecureBio.

Alliance Map

These run at the same time. One nonprofit board is the controlling stockholder of the lab, the committee that can halt its releases, a funder of one of the labs’ outside evaluators, the employer of OpenAI’s former VP of Global Impact, and, through one director, a bank chief whose institution also lists the committee chair as an advisor. The Foundation or its counterparties disclose each tie. The overlap is the finding.

Reach Assessment

Institutional: The structure page puts the nonprofit at the top of a company in which it holds 26 percent, and the Attorneys General’s documents give its safety committee a halt power over that company’s releases. Memetic: Its own public line is that safety is “only one part of the puzzle” and that a resilience ecosystem of independent institutions has to be built beside it.

Civilizational: OpenAI values the Foundation’s stake at “approximately $130B,” which Taylor calls “one of the best resourced philanthropic organizations ever,” against a stated $25B commitment and at least $1 billion of spending in the first year. What the record cannot show is how much of the evaluation field will come to rest on that money; the Foundation’s own AI Resilience grants were “to be shared publicly soon” at the time of its 1 June 2026 post.

Score note. Attack 9, defense 4, hp 9 on first-hand receipts: the structure page’s control and equity terms, the Delaware statement’s committee powers, California’s memorandum on the Class N stock and the services agreement, and the Foundation’s own board announcements. Peers: OpenAI (9/8/9, hazard 88) holds the same reach and entrenchment and a higher insulation because a commercial company’s money is its own customers’; the Frontier Model Forum (9/6/7, hazard 79) has the same reach and takes its money from the member firms it standardises; Open Philanthropy (8/8/9, hazard 80) is the field’s largest funder and is funded by a private family, not by the graded company; Schmidt Sciences (6/8/7, hazard 70) has less reach and a funder independent of the graded. The Foundation’s endowment is the graded company’s stock, which is the lowest insulation among them. Hazard 74 sits under the Forum and Open Philanthropy because its reach is one company, not a field, and above Schmidt Sciences.

Score basis. Atk 9 is reach: OpenAI’s structure page says the Foundation “appoints all members of the board of directors of OpenAI Group and can replace directors at any time,” and the Delaware statement records the committee’s “power and authority to require mitigation measures—up to and including halting the release of models or AI systems,” so the company’s product decisions on safety run through one nonprofit committee (OpenAI, Our structure; Delaware Department of Justice). The point under 10 is that the power sits inside one company and no instance of its use is on the record the page cites. Def 4 is funding insulation: the Foundation holds “a 26% equity stake in OpenAI Group, worth approximately $130B,” its board includes the chief executive of the company it oversees, and SecureBio’s Detection Director writes that the two “have almost identical boards”; the conditions that the committee stays in the nonprofit and the company’s directors “consider only the mission” on safety are what keep it above 3 (OpenAI, Our structure; Kaufman; Delaware Department of Justice). Hp 9 is entrenchment: the Class N stock, the services agreement that “will not be terminated without the NFP’s written consent,” the 21-day notice to the Attorney General and the relinquishment decision that “is in the control of the NFP Board” make it removable only by the Foundation itself and two state offices; the point under 10 is that the structure rests on a memorandum and a letter, not a statute (California Department of Justice, MOU). Peers on the same values: OpenAI at 9/8/9 and Open Philanthropy at 8/8/9 hold the same entrenchment; the insulation of 4 is the lowest of the four named peers because the endowment is the graded company’s equity.

Sources: OpenAI Foundation; Paul Christiano joins OpenAI Foundation Board — OpenAI Foundation (9 Sep 2026); Update on the OpenAI Foundation (24 Mar 2026); AI for Alzheimer’s — OpenAI Foundation (8 Apr 2026); Economic Futures in the Age of AI — OpenAI Foundation (27 May 2026); Resilience in the age of AI — OpenAI Foundation (1 Jun 2026); David Vélez and Robin Vince join OpenAI’s boards (21 Jul 2026); Supporting Communities: 2026 People-First AI Fund Grantees (1 Oct 2026); OpenAI — Our structure; Built to Benefit Everyone — Bret Taylor (28 Oct 2025); AG Jennings completes review of OpenAI recapitalization — Delaware Department of Justice (28 Oct 2025); California Attorney General MOU with OpenAI (27 Oct 2025); Building a Three-Day Early-Warning System for Novel Pathogens — SecureBio (21 Aug 2026); SecureBio — GPT-6 Astra assessment; Thoughts on Taking OpenAI Foundation Funding — Jeff Kaufman; Zico Kolter — Bio; Musk v. Altman, order 580 (20 May 2026); OpenAI, Inc. — ProPublica Nonprofit Explorer.