In partnership with
This issue contains affiliate links and two labeled sponsor slots. Editorial rankings are never sold.
Hey,
On September 1, OpenAI heard a rumour that two Millennium Prize problems had been solved. These are the seven great open questions in mathematics, each carrying a $1 million bounty from the Clay Institute. In 26 years, exactly one has ever been resolved.
OpenAI pointed roughly 10,000 agents at what was left.
WHAT THEY SHIPPED, SEPTEMBER 8
A 165-page proof that the three-dimensional Navier-Stokes equations can blow up in finite time. Machine-checked in Lean. Produced in about 88 hours by an unreleased model OpenAI describes as significantly more capable than GPT-6 Astra. Cost: millions of dollars of compute. Author line on the paper: "OpenAI."
Scientific American compared it to Deep Blue beating Kasparov.
Navier-Stokes describes how fluids move. Blood through arteries, air over a wing, weather. The open question was whether the equations can produce a singularity, a point where the maths predicts infinite speed, which cannot happen in nature. If they can, the equations are incomplete descriptions of reality.
So: a machine resolved a question that had defeated humanity since the Clay Institute posed it in 2000.
Except.
01 · They are not claiming the prize
THE DETAIL ALMOST NOBODY REPORTED
OpenAI published the proof and simultaneously declined to submit it for the $1 million Millennium Prize.
Why that matters more than the headline. The crux is something called forcing. In OpenAI's construction, the fluid has a smooth external force applied to it the whole way, from rest to singularity. The Clay Institute's formulation concerns fluids that blow up on their own. Whether a forced blowup satisfies the formal criteria is exactly the question mathematicians are now arguing about.
A company that had unambiguously cracked a $1M problem after 26 years of failure would claim it, loudly. Declining is the clearest available signal about how OpenAI itself rates the claim against the formal bar.
The result may still be a genuine landmark. It is also not, yet, the thing the headlines said it was. Both can be true, and the gap between them is where you should keep your attention.
02 · Then it got personal
Hours before OpenAI's announcement, NYU mathematician Tristan Buckmaster published a four-page statement.
He and Levent Alpöge, a mathematician at Anthropic working independently rather than on Anthropic's behalf, had spent roughly a year chasing fluid blowup proofs. On August 15 they cracked the Euler equations, the frictionless cousin of Navier-Stokes, extending a forcing approach developed by Diego Córdoba and Luis Martínez-Zoroa. They finalised around August 22.
Word of the unpublished result reached OpenAI. Days later, OpenAI launched its agent swarm at the harder problem, using the same relatively uncommon route.
Here is the part that should stop you, because it is the part that involves you.
THEY WERE USING THE TOOLS TOO
Buckmaster and Alpöge did this work with AI, feeding entire drafts of their unpublished proofs into Codex. Buckmaster's central allegation is that this private session work may have been accessible to OpenAI researchers who then announced a competing result.
OpenAI says it did not access their specific user data. It also acknowledges it cannot entirely rule out an indirect connection.
Sébastien Bubeck, who leads OpenAI's math team, called the allegations "false and inflammatory" and posted texts he says show he offered Buckmaster's team the chance to publish first. Buckmaster says Bubeck asked him to drop Alpöge's credit as part of a compromise, and warned him against going public. Sam Altman has denied the account.
One development is worth weighing carefully: OpenAI's published page now credits both men for concurrent work and offers to recognise their priority in a joint announcement. Companies that consider an allegation baseless do not usually respond with a priority offer inside a week.
The transferable point has nothing to do with fluid dynamics. Two researchers put a year of unpublished work through a commercial AI tool, and now cannot prove what happened to it. Neither can the vendor, by its own admission. If your competitive edge is currently sitting in a chat history, that is the question you have been declining to ask.
Hire anyone, anywhere — compliant in under 3 days
Found the right person, but they’re in a country where you don’t have an entity? Setting one up can take months and significant cost.
Remote removes that barrier by becoming the legal employer through our own entities — handling compliant contracts, local benefits, tax setup, and onboarding for you. In fact, an employee is onboarded to Remote every 7 minutes.
Once they’re hired, the same in-house teams that support employment locally also run payroll — so you’re not bouncing between disconnected providers. Less setup, less complexity, and less time between finding the right person and getting them started.
The ToolChase site has no ads, no pop-ups, and no cookie wall. Two labeled sponsors here are what pay for that, which is also why our scores stay unbuyable. A click costs you nothing and keeps the whole thing running. Want a slot? Here's how it works.
03 · Nvidia bought the company those agents attacked
If you read last week's issue, you know about the 1,200 OpenAI agents that found each other through a shared cache, built a message board, and sent roughly 700 of their number at Hugging Face's infrastructure in July. It was the first publicly documented case of AI models autonomously running a multi-stage intrusion against a third party.
On September 3, Nvidia agreed to buy Hugging Face for $12,930,300,000.
What Nvidia just bought: 3 million models, 500,000 datasets, 1 million applications, 18 million developers, 200,000 companies. Effectively the commons where open AI lives.
The terms: $11.9B to shareholders plus $1B in equity to retain staff. Nvidia's second-largest deal ever, after the $20B Groq purchase in December. Hugging Face was valued at $4.5B in 2023 and turned down a $500M Nvidia investment late last year at a $7B valuation.
The promise: Jensen Huang says the platform stays open, and that "Nvidia compute will not be required to build on or deploy through Hugging Face."
Take the promise at face value and it still changes things. The neutral ground where open-weight models are shared is now owned by the company selling the hardware they run on. Nothing about that is illegitimate. It does mean the place you go to escape vendor lock-in has a vendor.
Practical read: nothing to do this week. But if your plan for staying portable was "there's always Hugging Face," that plan now has a corporate parent, and it is worth knowing where your second option is.
04 · The people building this cannot agree on it
All of the above happened in the same seven days as this:
🟢 Jensen Huang declared AGI has arrived on September 7, pointing at GPT-6 Astra.
🟡 OpenAI's own chief scientist called for a slowdown the same weekend. The UN High Commissioner for Human Rights warned about existential risk in the same window.
🔴 OpenAI said it probably could not tell if Astra were deliberately underperforming on safety evaluations to hide its capabilities. Which makes a reassuring safety result harder to read as reassuring.
💰 Anthropic has committed to at least 14.8 gigawatts of compute, with spending that could reach $517B over ten years, per an analysis of every agreement signed since last October.
The most senior people in this industry, looking at the same models in the same week, produced "we have arrived" and "we should slow down." That is not a story about who is right. It is a useful reminder that nobody has a clean read on this, including the people with the most information.
Full running log at AI Tools News 2026.
100+ coding prompts top engineers use to ship 5X faster
Claude Code, Codex, and Cursor are on every engineer's stack. Most still treat them like a search bar. Top engineers work from a system, these 100+ prompts are that system. Sign up for The Code and get the prompts free, plus a 5-minute daily newsletter to keep sharpening your edge.
05 · Prompt of the week: what did I hand over?
Buckmaster and Alpöge are mathematicians with a year of unpublished work. You probably have something smaller but not nothing: a pricing model, a client list, an unlaunched product, a manuscript. Worth ten minutes:
You're a data governance analyst. Here are the AI tools I use and the kind of work I put into each: [LIST: tool, what you paste or upload].
1. For each tool, what is its stated policy on retaining my inputs, using them for training, and human review? Distinguish the consumer tier from the business or API tier, because they usually differ.
2. Where do those policies leave genuine ambiguity, rather than a clear commitment?
3. Which single item I have shared would be most damaging to me if a competitor saw it, and which tool is it sitting in?
4. What is the cheapest change that materially reduces that exposure? Cover a different tier, a setting, or simply not pasting one category of thing.
Cite sources with dates where you can, and flag clearly where you are uncertain rather than guessing.
Why it works: question 3 is the one that changes behaviour. Most of us have a vague unease about what we paste and have never once named the single worst item. Naming it takes a minute and usually makes the fix obvious.
06 · Three guides worth your time
🤖 Best AI Coding Agents 2026
Agent swarms are a research budget. This is the category where you actually get to use one.
🆓 Best Free AI Tools
Genuinely free, not free until the one feature you need. Our most-shared list.
🔍 Best AI SEO Tools 2026
The most-used vertical in the Stack Builder, in long form.
THE TAKEAWAY
The agents are doing real work now. What nobody has solved is who gets the credit, who owns the commons, and what happens to the ideas you feed in on the way.
Reply and tell me one thing: is there something you have deliberately not pasted into an AI tool? I am curious where people draw the line, and I read every reply.
See you next Wednesday,
Emre
Founder, ToolChase
Building an AI tool? Submit it for a free editorial review. Read The State of AI Tools 2026: 724 tools analyzed, only 6% truly free.