A New York University mathematician is accusing OpenAI of playing dirty in the race to crack one of mathematics’ most valuable unsolved problems.

Here’s what happened. An NYU researcher mentioned they were working on the Navier-Stokes existence and smoothness problem, one of seven Millennium Prize Problems with a $1 million bounty. Within hours, OpenAI reportedly redirected significant compute resources to solve it first.

The researcher calls it fighting dirty. OpenAI presumably sees it as fair competition. But the incident highlights a bigger shift in how mathematical research works when AI labs can throw nearly unlimited compute at problems the moment they hear about them.

The Non-Renewable Mining Problem

Terence Tao, one of the world’s leading mathematicians, has been warning about exactly this scenario. In a post on Mathstodon, he describes what’s happening as “non-renewable mining” of good mathematical problems.

The core issue: math research used to take months or years of sustained work. Researchers could share preliminary ideas, get feedback, and develop their approach. Now, even mentioning that you’re working on something promising can trigger a massive AI-powered effort to solve it before your research reaches its full potential.

Tao writes: “We have now seen that even the rumor of someone working on a problem can trigger a massive amount of AI-powered effort to flatten it before the original research project has time to reach its full potential.”

His conclusion is stark. The incentives may now point toward “no longer sharing any promising research leads, even within the research community.”

That’s a significant shift for mathematics, which has historically been unusually open and collaborative compared to other fields.

What This Means for Research

If Tao is right, we’re watching a fundamental change in how mathematical research gets done.

The old model relied on researchers sharing work-in-progress. You’d present a promising direction at a conference, get feedback, refine your approach. The community benefited from seeing how ideas developed.

The new model might look more like stealth mode startups. Don’t talk about what you’re working on until you’ve solved it. Don’t share preliminary results. Keep promising leads secret.

That’s bad for mathematics as a field. Progress accelerates when people can build on each other’s ideas. Collaboration beats isolation, usually.

But it’s rational behavior for individual researchers. If mentioning your research direction means a well-funded AI lab will throw enormous compute at the problem and potentially scoop you, why would you share anything?

The Bigger Picture

This isn’t just about one million-dollar prize problem. It’s about how AI changes the economics of research.

When solving hard problems required human insight developed over months or years, sharing preliminary work carried manageable risk. By the time someone else understood your approach well enough to apply it, you’d made more progress.

When solving hard problems can be partially automated through massive compute, sharing preliminary work means handing your competitors a roadmap. They can potentially execute on your ideas faster than you can, especially if they have more GPUs.

The NYU mathematician’s complaint isn’t really about fairness in some abstract sense. It’s about whether individual researchers can still make careers by working on hard problems when AI labs can respond to rumors with overwhelming computational force.

What OpenAI Is Actually Doing

To be clear, we don’t have OpenAI’s side of this story in detail. The company hasn’t publicly commented on the specific accusation.

What we do know: OpenAI has been showcasing GPT-5.6 Sol’s ability to help with advanced research. They published a case study about an MIT researcher using it to autonomously run quantum computing experiments, analyze results, and calibrate qubits.

The message is clear. These models aren’t just for coding productivity anymore. They’re research tools that can participate in actual scientific work.

That’s legitimately impressive. It’s also what makes situations like the Navier-Stokes incident fraught. When your AI can meaningfully contribute to solving million-dollar math problems, and someone mentions they’re working on one, what do you do with that information?

Who Should Care

If you’re doing research in mathematics, theoretical computer science, or related fields, this matters a lot. The ground rules are changing.

If you’re building AI tools for research, think about the incentives you’re creating. Making it easy to throw compute at problems the moment you hear about them might not be the healthiest dynamic for research communities.

If you’re neither of those things, this is still worth understanding. We’re watching a live experiment in how AI changes knowledge work. The same dynamics playing out in mathematics research will likely show up in other fields where problems can be parallelized across enough compute.

The question isn’t whether AI will be useful for research. It clearly is. The question is whether the transition creates incentives that make research communities more closed, more competitive, and less collaborative.

Based on what Terence Tao is seeing in mathematics, the early signs aren’t great.