Here’s what happened this week: OpenAI announced it had solved one of the seven Millennium Prize Problems in mathematics, a 90-year-old puzzle called Navier-Stokes that carries a million-dollar bounty. Impressive, right? Except according to Tristan Buckmaster, a math professor at NYU, OpenAI knew his collaborator was working on the same problem and raced to beat him to publication anyway.

That’s not just bad form. It’s a fundamental break in how scientific research is supposed to work.

The details matter here. Buckmaster says his collaborator, Levent Alpöge, had been making progress on the problem. Word got out, as it does in academic circles. And then, apparently, OpenAI threw massive computational resources at flattening the problem before Alpöge could finish his work. We’re talking 10,000 concurrent AI agents and an unreleased model more powerful than GPT-6.

OpenAI got its headline. Alpöge got scooped on what could have been a career-defining achievement.

The Tragedy of the Mathematical Commons

Terence Tao saw this coming. Tao, widely considered one of the greatest living mathematicians, wrote recently about how “the collection of good, fruitful open problems is now being mined in a non-renewable fashion.” He warned that even rumors of someone working on a problem could “trigger a massive amount of AI-powered effort to flatten it before the original research project has time to reach its full potential.”

That’s exactly what appears to have happened here.

Think about the incentives this creates. Tao puts it bluntly: “The incentives may now be pointing in the direction of no longer sharing any promising research directions until after one has published.”

This is catastrophic for mathematics and science generally. Progress depends on researchers sharing ideas, building on each other’s work, and yes, sometimes competing. But there’s supposed to be a baseline of professional conduct. You don’t weaponize unlimited computational resources to snipe someone else’s years of work just because you can.

This Isn’t About One Problem

The Navier-Stokes controversy is part of a pattern. Gary Marcus has been documenting what he calls OpenAI’s “egregious pattern of misconduct” — nine troubling reports in seven days, by his count. The details vary, but the theme is consistent: move fast, claim credit, worry about the consequences later.

And it’s not just OpenAI. The same day this news broke, an Anthropic safety researcher said there’s more than a 10% chance AI “could kill all humans” by the end of the decade. His colleague Jacob Coxon quit over concerns that Anthropic and its rivals are “carelessly racing to build superhuman systems they cannot control.”

These stories are connected. They’re both about what happens when the incentive structure is “ship it and see what breaks.”

With the Navier-Stokes problem, what breaks is the collaborative infrastructure of mathematical research. With AI safety, what breaks might be considerably worse.

The Real Cost

There are only seven Millennium Prize Problems. They’re called that because they’re supposed to be hard enough to occupy the best mathematical minds for generations. They’re not meant to be computational target practice for frontier AI labs looking for impressive demos.

Even if OpenAI’s solution is completely valid (and that’s still being verified), the way they got there matters. Buckmaster isn’t just upset about being scooped. He’s pointing out that this fundamentally changes how mathematical research works. And not for the better.

The argument from the AI labs will be predictable: this is progress, this is what these tools are for, mathematics is better off with these problems solved. But that misses the point entirely. Mathematical progress isn’t just about having answers. It’s about the process of getting there, the insights discovered along the way, the community of researchers advancing human understanding together.

Strip-mining that for PR victories is a choice. It’s not an inevitable consequence of AI progress. It’s a decision by specific people at specific companies to prioritize speed and spectacle over the health of the research ecosystem.

What Happens Next

If Tao is right, we’re about to see mathematics become dramatically more secretive. Researchers will stop sharing promising directions until after publication. Collaboration will suffer. Progress will slow in ways that are hard to measure but very real.

And the AI labs will keep racking up solved problems and claiming credit for advancing human knowledge, even as they make the actual practice of advancing human knowledge harder for everyone else.

This is a choice point. We can decide that this kind of behavior is unacceptable, that there should be norms and consequences when AI labs burn through shared scientific resources for competitive advantage. Or we can shrug and accept that everything is fair game now, that unlimited compute beats careful research, and that the primary value of unsolved problems is their PR potential when solved.

I know which future I want. The question is whether anyone with actual power over these companies feels the same way.