The 130 Billion Dollar Question About the Co-Founder Who Left OpenAI to Watch OpenAI

Wojciech Zaremba spent a decade building the systems that made OpenAI the most watched AI lab on earth. Now he has a new job: watching OpenAI.

Zaremba, one of OpenAI's least public co-founders, announced Monday that he is leaving frontier research to run AI resilience at the OpenAI Foundation — the nonprofit arm that controls the company it is supposed to hold accountable. The foundation published its AI resilience framework Monday, laying out a $130 million initial grant slate across biosecurity, cybersecurity, model safety, and AI's effects on children, with a long-range pledge of $25 billion.

The numbers are real. The structure is the problem.

The OpenAI Foundation holds a 26 percent stake in the for-profit OpenAI Group, currently valued at roughly $130 billion — one of the largest nonprofit endowments in American history, per its own announcements. It pledged $25 billion across AI resilience and life sciences. But in 2024, according to tax filings reported by the Associated Press, the foundation received $4,433 in public donations and deployed approximately $50 million. At that rate, reaching the $25 billion pledge would take five thousand years.

The governance compounds the math problem. Of the foundation's nine board members, only two have meaningful nonprofit experience. Four — including Bret Taylor and Sam Altman — hold direct financial interests in OpenAI stock appreciation, according to the Longterm Wiki, a research organization focused on AI governance. That means the people tasked with ensuring the foundation acts independently of the company have a personal financial stake in the company's valuation going up.

Catherine Bracy, founder of the equity nonprofit TechEquity and a member of the EyesOnOpenAI coalition, put it plainly. "The unspoken truth here is that they're never going to make a decision that is bad for the company," she told Vox. "These two entities cannot live under the same roof where the mission is in control."

The foundation's independence from OpenAI the company is not merely philosophical. The entity approved the for-profit restructuring last fall over objections from California Attorney General Rob Bonta, a coalition of nonprofits, and Elon Musk, who is currently in trial arguing that OpenAI's conversion violated its charitable mission. The company has also faced scrutiny for a Pentagon defense contract, lobbying against statewide AI safety legislation, and deploying ads on its free ChatGPT tier — decisions the foundation has not publicly opposed.

Zaremba, in his resilience essay, drew an analogy. Fire could not be aligned, he wrote — only made safer through resilient infrastructure. Electricity followed the same path. We need to be resilient to the outcomes, he wrote.

The analogy has an uncomfortable edge. The infrastructure that made electricity safe — independent testing bodies, industry standards, public codes — was built by institutions with no financial stake in whether the electrical grid succeeded. The OpenAI Foundation's board has a $130 billion financial stake in whether OpenAI succeeds.

LASST, a legal advocacy organization focused on AI safety, flagged the same tension in its analysis of the foundation's 2026 spending plans. "Will the Foundation fund evaluators who can be genuinely critical of OpenAI's own models?" the group asked. "Will it push for standards that impose real constraints?" The answers matter, LASST concluded — and the foundation's track record so far offers little evidence either way.

Zaremba did not respond to a request for comment. The foundation declined to comment beyond its published statements.

The grant slate announced Monday — $130 million across four areas — is real money by ordinary standards. It is less than one percent of the foundation's paper wealth. Whether it represents the beginning of serious deployment or a structural ceiling depends entirely on whether the foundation can act against its own company's interests. The governance suggests it cannot. The track record suggests it has not had to prove otherwise yet.

The test will not be the grants the foundation issues to organizations that agree with OpenAI. It will be whether Zaremba — or anyone on that board — is willing to fund research, standards, or evaluations that OpenAI the company opposes. That has not happened. And the structure may make it impossible.