OpenAI Reverses Course: Nonprofit Keeps Control of Firm

OpenAI has backtracked on one of the most contentious plans in its recent history. The company had proposed turning its commercial arm into an independent entity, free from the oversight of the nonprofit foundation that has governed it since its founding. That plan is now off the table. As OpenAI itself announced, the nonprofit will continue to control the company, just as it has from the start.

The shift is far from cosmetic. For months, OpenAI argued it needed a more conventional structure to attract the capital required to compete in the race toward artificial general intelligence (AGI). The original proposal would have diminished the nonprofit's influence in favor of a more autonomous for-profit subsidiary. That idea drew a wave of criticism from former employees, academics and, decisively, from the attorneys general of Delaware and California — the two states with jurisdiction over OpenAI's legal structure.

What Actually Changes

The plan that is moving forward keeps the nonprofit in the driver's seat, but converts the commercial subsidiary — the LLC that has operated under the nonprofit's umbrella since 2019 — into a Public Benefit Corporation (PBC). It's a corporate structure designed for companies with a stated purpose, legally required to weigh shareholder interests alongside their broader mission. OpenAI notes that this structure is already standard at other AI labs, including Anthropic and X.ai, and is also used by socially minded companies like Patagonia.

The nonprofit won't just retain control — it will also become a major shareholder in the PBC, at a stake to be determined by independent financial advisers. According to the company, this will give the nonprofit greater resources to fund programs aligned with its mission, resources that will grow as the business grows.

OpenAI is also scrapping its old "capped-profit" structure, a mechanism that placed a ceiling on investor returns. The company explains that design made sense when it looked like there would be only one dominant player in the race to AGI, but no longer fits a landscape with several major labs competing. In its place, OpenAI is adopting a conventional equity structure, with shares for everyone involved. The company insists this isn't a sale, but a change in legal form.

Sam Altman's Take

CEO Sam Altman sent a letter to employees and other stakeholders explaining the reversal. In it, he recalls that OpenAI "is not a normal company and never will be," and revisits the company's origins: a group of people gathered around a kitchen table, with no business plan and no clear sense of how artificial general intelligence would even get built.

Altman describes a tension that has run through OpenAI's entire history: early on, many inside the organization believed AI should remain in the hands of a small group of trusted people who could "handle it." Today, he writes, they see a different path — one where AGI directly empowers every individual as the most capable tool ever created. He acknowledges not all of that use will be positive, but says he trusts the benefits will outweigh the harms "by orders of magnitude."

That bet is what Altman calls "democratic AI": powerful tools in everyone's hands, open models, and broad freedom for users to decide how they use ChatGPT within general limits. He points to people using ChatGPT to boost their productivity as scientists or programmers, to tackle health problems, or to seek advice in difficult situations. He also admits a current constraint: demand outstrips what OpenAI can supply, forcing the company to keep usage limits in place.

Three Goals and a Geopolitical Warning

The letter lays out three goals for this new chapter. The first is to be able to operate and raise enough resources to make its services broadly accessible to all of humanity — something that today requires hundreds of billions of dollars and, Altman says, could eventually require trillions. The second is for the nonprofit to become the largest and most effective nonprofit in history, focused on using AI to deliver the highest-impact outcomes for people. The third is to deliver beneficial AGI, built on the company's safety and alignment track record: red-teaming processes (adversarial testing to catch flaws) and transparency about model behavior through what it calls the "model spec."

Altman frames all of this within a larger struggle: he wants "democratic AI" to win out over "authoritarian AI," placing the debate over OpenAI's governance squarely within the geopolitical competition over AI development.

What Comes Next

OpenAI says it will keep working out the details of the plan with the offices of the attorneys general of Delaware and California, with Microsoft, and with the commissioners it recently appointed to the nonprofit's board. Those commissioners, the company says, will soon put forward recommendations on how the nonprofit can support a future of "more democratic AI," with an impact on areas like health care, education, public services and scientific research.

The episode offers a lesson on the real limits of corporate power at OpenAI. The most closely watched company in the AI industry set out to rewrite its own governance to make fundraising easier, and had to retreat under combined pressure from internal voices, academics and state regulators. The nonprofit is still calling the shots. What remains to be seen is how much that control will weigh on the decisions that shape the development of artificial general intelligence in the years ahead.