The Mother of All Mission Drifts

There is a document you have probably never read. It is called a Form 990 — the annual tax filing that American nonprofits are required to submit to the Internal Revenue Service. It is dense, bureaucratic, and almost entirely unread by anyone outside the accounting profession.

In November 2025, buried inside OpenAI's final Form 990 — the last one it would ever file as a nonprofit — a single word had been removed from the company's mission statement. The word was safely.

Nobody held a press conference. There was no blog post. The deletion surfaced in a footnote of a corporate restructuring document that most people would never read. ^[1] But it was noticed — and it became shorthand for something that had been building at OpenAI for years: the systematic subordination of safety to speed, of caution to capital, of the founding vision to the demands of a company growing faster than any technology enterprise in history.

For anyone evaluating which AI companies to trust with critical infrastructure — in healthcare, in government, in financial systems — this is not a footnote. It is the entire story.

How it started

OpenAI was incorporated in December 2015 as a nonprofit research laboratory. Its founding mission, as filed with the IRS, was explicit: to build general-purpose artificial intelligence that safely benefits humanity, unconstrained by a need to generate financial return.

The founding coalition — Sam Altman, Greg Brockman, Ilya Sutskever, Elon Musk, and others — made a specific argument: that the development of artificial general intelligence was too consequential to be left to for-profit companies whose incentives would inevitably compromise safety. The nonprofit structure was not incidental. It was the point. ^[2]

The safety exodus

The deletion of a word from a tax filing is a symbol. The substance arrived earlier, and more visibly, in the form of departures.

In May 2024, Ilya Sutskever — OpenAI's co-founder and chief scientist, the person most identified with its original research mission — announced he was leaving. Hours later, Jan Leike, co-lead of the company's superalignment team, also resigned. ^[3]

"Over the past years, safety culture and processes have taken a backseat to shiny products. We are long overdue in getting incredibly serious about the implications of AGI."

— Jan Leike, former co-lead, OpenAI Superalignment team · May 2024

Leike's departure was notable for another reason: in making it public, he was almost certainly forfeiting millions of dollars in vested equity. OpenAI's nondisparagement agreements are notoriously strict. He said it anyway.

They were not alone. Leopold Aschenbrenner had been fired months earlier for raising information security concerns. Daniel Kokotajlo, William Saunders, and others had already left. The Superalignment team — announced in 2023 with a commitment of 20% of the company's computing power — was dissolved days after Sutskever and Leike's departures. One year after it was created. ^[4]

The trial that didn't answer the question

In May 2026, Elon Musk's lawsuit against OpenAI went to a three-week federal trial in Oakland. The core claim was simple: that Sam Altman and Greg Brockman had betrayed OpenAI's founding mission, and that Musk's $38 million in seed funding had been given on the explicit condition the company would remain nonprofit and open source.

The jury took less than two hours to decide. Not on the merits. Musk had filed outside the statute of limitations. ^[5]

What the trial did not do — what no court has yet done — is rule on whether OpenAI's for-profit conversion was a betrayal of its founding mission. The question that matters most remains unanswered in any formal sense, which means the answer is left to the market.

What this means for Gulf buyers

The Gulf is building AI infrastructure at sovereign scale. Saudi Arabia's Project Transcendence, the UAE's national AI strategy, Qatar's ambitions in financial AI, Bahrain's healthcare digitization — these are not consumer applications. They are decisions about which systems to trust with national data, public health records, government services, and critical infrastructure. The time horizons are decades, not quarters.

For buyers operating at that scale, the OpenAI trajectory raises a specific and practical question: when a company changes its mission, what happens to the commitments it made before the change?

This is not an abstract ethics question. It is a procurement question. A governance question. A question about which companies are structurally capable of maintaining the commitments they make — and which are subject to the same incentive drift that reshaped OpenAI.

The structural argument

The difference between OpenAI and Anthropic is not primarily a difference in capability. Both companies build frontier models. Both have significant commercial operations. Both compete in the same enterprise AI market.

The difference is structural. Anthropic was built from the outset with a Long-Term Benefit Trust holding a controlling interest — specifically designed to prevent the kind of investor capture that reshaped OpenAI. The mission is not a stated aspiration. It is embedded in the governance architecture.

Several of Anthropic's founders left OpenAI precisely because of the tension Leike described publicly. Dario and Daniela Amodei, along with senior researchers, departed in 2021 — three years before the Superalignment exodus — citing concerns about safety and direction. They did not write blog posts. They built a company structured to resist the incentives they had watched distort the one they left.

That history is not a marketing message. It is a design specification. And in markets where long-term trust is the basis of the relationship — which describes every sovereign and enterprise AI deployment in the Gulf — design specifications matter more than sales decks.

The question for the region

At Qlabs Intelligence, we work with institutions navigating exactly this evaluation. The Gulf is not short of AI interest. It is short of frameworks for distinguishing between AI companies whose commitments are structural and those whose commitments are contingent on the capital environment not changing.

OpenAI's trajectory is the most documented case study available on what happens when a mission-driven AI company meets the incentives of frontier model development at scale. The answer — for now — is that capital wins. The mission is rewritten. The safety researchers leave. The word is deleted from the filing.

The Gulf deserves AI partners who have built the structures to make that outcome harder. The evaluation of which companies those are — and what presence they should have in the region — is exactly the work this series is doing.