Ninety years. That’s how long the Navier–Stokes problem sat there, unmoved, before OpenAI announced in September 2026 that one of its AI systems had resolved key parts of it. Not the whole thing. Parts. And in the same stretch of time, the company also stood up an advisory council on wellbeing and AI, plus a separate safety advisory group for generative AI.
So we have one number that sounds like history being made, and a pile of committees that sound like a company preparing for questions it does not want to answer.
What was actually claimed
Per reporting from The Business Standard on 9 September 2026, OpenAI says its AI system resolved key parts of the 90-year-old Navier–Stokes math problem. That phrasing matters, and I want to sit on it for a second, because the gap between “resolved key parts of” and “solved” is where most of the hype in this industry lives.
Navier–Stokes describes how fluids move. Air over a wing, blood through an artery, water down a drain. The equations work in practice, which is why engineers have used them for a century. What nobody could prove is whether smooth solutions always exist and stay smooth, or whether the math eventually blows up. That proof is one of the hardest open questions in mathematics.
Resolving parts of it is real work. It is also the kind of claim that no press release can fully validate on its own. Mathematics has a specific verification process, and it is slow, adversarial, and conducted by people who are professionally motivated to find your mistake.
Why the advisory group is the more interesting story
OpenAI formed a math advisory group. It also formed an advisory council on wellbeing and AI. It also created a safety advisory group for generative AI, alongside granting its board veto power over certain decisions, as Voicebot.ai reported.
Notice the pattern. Each time OpenAI does something that draws scrutiny, an advisory body appears shortly after. I am not saying that is cynical by default. Bringing in outside expertise is the correct move when you are making claims in a field you do not own. Mathematicians should be checking mathematical output. Clinicians and researchers should be weighing in on wellbeing questions.
But advisory groups have a structural weakness that I have watched play out repeatedly across this industry: they advise. They do not decide. The coverage of the wellbeing council asked exactly the right question, whether the company will bother listening to those experts. That question applies just as much to a math advisory group. If the advisors flag that a result is overstated, does the announcement get walked back? Does the blog post get edited? Or does the group exist so that future announcements come pre-blessed?
Those are different products. One is quality control. The other is credibility laundering. From the outside, on day one, they look identical.
The misconduct headline nobody should skip past
There is also a headline in circulation arguing that OpenAI’s latest math results commit research misconduct. I have not seen the substance of that argument verified, and I am not going to pretend I have. But its existence tells you something about how the announcement landed with the people best equipped to evaluate it.
When a lab announces a mathematical result and the response from parts of the research community is an accusation about process rather than a debate about content, that is a signal about trust. Not about whether the math is right. About whether anyone believes the framing around it.
What I would want before calling this a milestone
- The actual write-up, in full, available to anyone who wants to attack it
- Named mathematicians outside OpenAI confirming which parts were resolved and which were not
- A clear statement of what remains open, written by someone with no equity in the company
- Evidence that the math advisory group can publicly disagree with OpenAI and keep its seat
None of that is unreasonable. All of it is standard in mathematics. The field already has a solid process for exactly this situation, and it does not require a corporate advisory structure to function.
My read
I think something real happened here. I also think the packaging is doing more work than the result. A 90-year-old problem, partially resolved, announced by the company that stands to benefit most from the headline, followed by the formation of a group to advise on the very thing already announced.
The sequence is backwards. You convene the experts, then you publish. OpenAI published, then convened.
If the math holds, it holds regardless of the press cycle, and mathematicians will confirm it on their own timeline. If it does not hold, no advisory council will save the announcement. Either way, the committees tell you what OpenAI expects to be asked. That is worth watching more closely than the equation.
🕒 Published: