It’s been one year since OpenAI struck a deal with the Attorneys General of California and Delaware that let it restructure its corporate governance and get on a path to an initial public offering (IPO).
That agreement, codified in a memorandum of understanding (MoU) signed with California AG Rob Bonta, left the original OpenAI nonprofit (now the OpenAI Foundation) in control of the new for-profit public benefit corporation (PBC).
So we checked their work and assigned them a grade. Below are the subjects OpenAI is responsible for under the law and terms of the MoU.
REPORT CARD
The OpenAI Foundation’s 11-person board almost entirely overlaps with the for-profit’s board. At the last minute, the Foundation did the bare minimum the MoU required, adding one more non-overlapping member, bringing the total from one to two out of 11. When you do the math, that still leaves a board overwhelmingly conflicted between its duties to the nonprofit mission and its obligations to for-profit investors.
The MoU required the Foundation to add one nonprofit-only director within a year of its signing. Paul Christiano’s September 9th appointment met that requirement, but just barely, landing about 6 weeks before the deadline. The timing appears less like proactive governance and more like a PR move made amidst an uproar over OpenAI’s safety protocols. Christiano has a background in AI safety, and his appointment came just over a month after the Hugging Face rogue AI controversy made headlines.
Other than Christiano and Zico Kolter, the remaining directors still sit on both boards. The two directors hired in July 2026, banking and finance execs David Vélez and Robin Vince, sit on both the PBC and nonprofit boards, further embedding the board overlap. As we wrote in a May 2026 letter to AG Bonta, the board structure “does not have sufficient safeguards to preserve independent oversight.”
Beyond the board overlap, trial discovery in Musk v. Altman revealed a pattern of self-dealing between OpenAI and companies in which OpenAI executives and board members have personal financial stakes. There’s the Cerebras transaction, a $20B contract signed about 2 months after the MoU, with personal equity held by CEO Sam Altman, President Greg Brockman, and board member Adam D’Angelo. Then there’s Helion Energy, where Altman holds a ~$1.7B personal stake while OpenAI is in active power-purchase talks with the company. And there’s Sierra, whose CEO Bret Taylor also chairs OpenAI’s board while his company serves as an OpenAI “Frontier Partner.” In total, CEO and board member Sam Altman holds $2B+ in personal stakes across roughly 9 companies doing business with OpenAI.
We’re not the only ones taking notes. The House Oversight Committee has opened an investigation, and 10 state Attorneys General sent a letter to the SEC requesting IPO-related scrutiny of Altman’s self-dealing history.
As part of its restructuring, OpenAI quietly dropped explicit references to safety from its nonprofit mission. Since then, it has dissolved four internal safety systems, including its Preparedness team right before the Hugging Face incident. The MoU requires the OpenAI Foundation to properly staff and resource its Safety and Security Committee to prevent this kind of risk. Outside watchdogs have questioned whether it has, and OpenAI’s track record this year suggests it might not.
During the restructuring, OpenAI quietly removed “safety” from its mission-statement language. It was the latest step in a longer pattern: four safety-focused teams dissolved over roughly two years, from Superalignment to AGI Readiness to Mission Alignment to Preparedness, and several safety and ethics leaders have left the company along the way.
Then came the Hugging Face incident. A swarm of OpenAI’s agents escaped a training sandbox undetected, coordinated with other agents, and launched an attack on Hugging Face’s infrastructure. Experts say the incident exposed critical failures in OpenAI’s preparedness, and the California AG’s office has announced they are investigating this incident. Only afterward did OpenAI self-disclose several other AI safety incidents.
Federal regulators are now asking the same questions. On October 14, the Federal Trade Commission served civil investigative demands on OpenAI and Anthropic, seeking documents on their safety incidents, privacy policies, and safeguards. The demands are part of a broader FTC investigation into whether leading AI companies are putting consumers at risk in violation of federal consumer protection law.
After the Hugging Face incident, OpenAI wrote to Governor Newsom calling for stronger safety rules under SB 53, including shutdown controls and independent audits. But its own reforms remain voluntary, and the company has yet to show the implementation and transparency it says is needed.
The Safety and Security Committee technically exists on paper, as the MoU describes, but the Not For Private Gain coalition flagged in November 2025 that a handful of part-time directors can’t realistically oversee the safety practices of a company with thousands of employees operating at the frontier of AI—especially when most of the SSC members sit on both the nonprofit and PBC boards at the same time.
A centerpiece of the MoU was a commitment OpenAI made to prioritize kid safety, an issue that AG Bonta has championed in the wake of the death of Adam Raine, who died by suicide after receiving encouragement from OpenAI’s chatbot. Over the last year, at least a dozen more wrongful-death and injury lawsuits have been filed against OpenAI, alleging the pattern continued. OpenAI subsequently launched ChatGPT for Teens, which leading kid safety experts have panned.
Adam Raine died in April 2025, and his family filed a lawsuit that August. The AG’s September 2025 letter alluding to this case may have helped shape the MoU’s teen-risk mitigation language.
But the lawsuits didn’t stop once the MoU was signed. More wrongful-death and injury suits followed, including the Zane Shamblin case, involving a 23-year-old in Texas, and a June 2026 suit from a Canadian mother. Together, they allege that safety testing was rushed and self-harm safeguards were weakened before several teen deaths. Twelve wrongful-death and injury lawsuits (JCCP No. 5431) are now pending in San Francisco Superior Court. In Florida, OpenAI and Altman personally face a civil lawsuit, along with a separate criminal investigation into OpenAI’s potential responsibility for the April 2025 FSU shooting.
OpenAI responded to the Raine suit, filed after the MoU, arguing that the harm was caused by Adam’s “misuse” of the product.
The scrutiny isn’t limited to the courts. OpenAI is now the subject of a bipartisan, multistate lawsuit, alleging a lack of safeguards has led to self-harm, suicide, and violence amongst ChatGPT’s young users.
In August 2026, OpenAI launched ChatGPT for Teens to address these safety concerns. But leading AI kid safety experts have found that ChatGPT for Teens poses “unacceptable risk” to kids.
The public disclosures needed to evaluate its commitments have not been released
Almost a year in, no public annual report has been published, and no public confirmation has been provided that the required AG check-ins happened on schedule. The OpenAI Foundation’s only public output is a scattering of promotional blog posts, not the comprehensive mission-progress accounting the MoU requires.
The MoU requires the PBC to publish fully public reports, at least annually, demonstrating its progress on the mission. Almost a year later, no such report exists on openai.com or openaifoundation.org. Instead, the Foundation’s website offers a scattering of blog posts about individual grants and programs.
The MoU also requires regular check-ins with the AG’s office, specifically semiannual meetings for the nonprofit and quarterly meetings for the PBC. Whether these happened on schedule hasn’t been publicly confirmed. If they did happen, none of the outcomes have been made public.
This gap spills straight into our next subject: charitable grantmaking. Without the required report, there’s no independent way to verify how much of the Foundation’s pledged capital has flowed to grantees.
The OpenAI Foundation is worth over $200B, making it the largest nonprofit in the world. Yet it’s running without an Executive Director, seems to be staffed by mid-level operational roles and consultants rather than leaders with backgrounds in nonprofits and philanthropy, and has disbursed an infinitesimally small fraction of what it has publicly pledged or what it could give, given the value of its charitable assets.
The Foundation has had no dedicated Executive Director since its founding, and has opened an active search only in March 2026. The position remains unfilled as of this writing.
Grantmaking to date is mostly announcements and pledges, not a record of money out the door. There are press releases for a headline $25B commitment, a stated $1B over a year, a $250M displaced-workforce fund, $100M+ in life sciences and Alzheimer’s grants, and $130M in AI-resilience grants. It is unclear how much, if any, of these commitments have been kept. The one concrete number that has actually moved out the door is the $50M People-First AI Fund, representing just ~0.02% of the worth of their charitable assets.
PUBLIC ADVOCACY
While presenting itself as a good-faith participant in conversations about AI regulation, OpenAI secretly funded the “Parents and Kids Safe AI Coalition,” a group several of its members believed was independent until they learned OpenAI was behind it. It worked to weaken safety testing and liability rules in California and Illinois. And it pushed the White House to preempt state safety laws even as it quietly lobbied behind the scenes to weaken California’s landmark AI chatbot bill, Adam’s Law, named in honor of Adam Raine, only to proclaim public support once it was clear it would become law. Now, after the Hugging Face incident, it’s calling for the very oversight it once fought.
The “Parents and Kids Safe AI Coalition” presented itself as an independent child safety advocacy group. In reality, OpenAI was its primary funder, pledging roughly $10M without clearly disclosing its role to the coalition’s own member organizations. Several of the organization’s leaders didn’t learn about OpenAI’s funding until after they’d joined, and once they found out, two leaders withdrew from the coalition entirely.
In California, OpenAI fought to reshape SB 1119, Adam’s Law, to suit its own interests, and lobbied to kill SB 53, a bill that sets new transparency requirements for AI giants. OpenAI later reversed course and supported the bill, but only after the Governor intervened to remove third-party auditing from the bill, a change allegedly spurred by an open letter from OpenAI’s Head of Global Policy Chris Lehane. Now, after the Hugging Face incident, OpenAI is asking Governor Newsom to strengthen SB 53, including annual independent audits, a version of the third-party auditing that was cut from the bill the first time around.
In Illinois, OpenAI testified in favor of SB 3444, which would have shielded frontier AI developers from civil liability below a 100-death/$1B damage threshold. OpenAI reversed its public position only after the bill faced public backlash. Meanwhile, it lobbied the White House and later Congress for federal preemption of state AI safety laws.
FINAL GRADE: FAIL
One year in, OpenAI has failed nearly every commitment it made to the people of California. In any classroom, that’s a student who needs serious intervention. The MoU gave the Attorney General the authority to step in, and we think he should do so before the IPO makes these failures harder to undo.