OpenAI is the company that put artificial intelligence in everyone's pocket through ChatGPT, the program where you type a message and get an answer as if you're talking to a person. Its story, though, has a double-bottomed twist: it started out with a promise to protect people from dangerous artificial intelligence, and today it's among those building it most aggressively. So it's worth knowing who's actually standing behind the screen.
OpenAI didn't start as a company. It started as a vow: a group of researchers promised they'd build artificial intelligence that serves everyone, not a handful of owners, and that profit wouldn't be the engine. So at first they registered as a nonprofit organization. That's a legal form that, by law, doesn't distribute profit to owners, but puts everything back into the mission.
The vow held up briefly. Building modern artificial intelligence needs money that donations don't cover: buildings full of machines computing around the clock. So OpenAI opened a side door for investors. Microsoft came in first, and with the biggest sum. Imagine calling in a fire inspector to impartially check how fire-safe the building is, and then he buys a stake in the construction firm putting it up, and becomes its partner. Whose interests will he protect going forward, the people inside the building, or his investment in it?
Then ChatGPT arrived and everything sped up. The money that flowed in after that was too much for the old, idealistic setup. Last year OpenAI finished the transition: the nonprofit now holds only a stake in the company, not all of it, though it keeps the right to choose the governing board, the company's board. Microsoft holds just as large a stake and stays a key partner for years ahead.
Picture a volunteer fire brigade, founded on an oath to save people from fire for free, for the common good. Over time it opens a construction division to pay the bills for equipment and gear. The construction division grows bigger than the fire brigade itself and takes on an outside partner. Formally the fire brigade still stands at the door and approves who's in command. But the money and the direction now come from the construction division.
On October 6, 2026 OpenAI published on GitHub a broad set of new mathematical results from one of its internal models, with many of the proofs also written in Lean - a language in which a computer checks every step. The company discussed how to release them with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study and drew on its recommendations. By OpenAI's numbers, the average result took compute equal to about three hours of ChatGPT Pro thinking. A month earlier, with the Navier-Stokes problem, there wasn't a single independent voice on the page. This time at least the way of releasing was discussed with people from outside, but whether the results hold is something mathematicians will say once they sit down with them.
Here's the weight of it
I don't think OpenAI is lying when it says it still cares about safety. I think the incentives no longer let them stop. Once you let an investor in through the side door, they want their money back, not a wait for perfect safety. And the excuse everyone reaches for is familiar: if they stop, someone else, more reckless, will keep going. That's the exact argument every race with something dangerous uses, right before it ends badly. I'm not saying I know how it ends. I'm saying the guardian is now running alongside everyone else, not ahead of them.