The seven questions are known collectively as the Millennium Prize Problems. They were named in 2000 by the Clay Mathematics Institute, which offered USD 1 million for a correct solution to each. The Navier-Stokes existence and smoothness problem is one of the seven, and it remains formally unsolved: no accepted proof has resolved it, and no official prize has been awarded. Claims that artificial intelligence has solved it should be read against that baseline.
The list exists because these problems sit at the load-bearing walls of modern mathematics. A proof or disproof would not merely settle a puzzle. It would change how physicists model fluids, how cryptographers trust codes, and how mathematicians understand the limits of computation itself. That is why the prize is unusual: the money is small compared with the status, and the status is small compared with the consequences.
Mechanism: why these seven and what each demands
The Clay Institute did not pick seven random hard problems. It picked problems whose resolution would reorganize entire fields. Each asks a yes-or-no question whose answer forces a structural insight, not a clever trick. The table below shows what each problem asks and why a solution would matter.
| Problem | Core question | Why a proof would matter |
|---|---|---|
| P versus NP | Can every problem whose solution is quickly checkable also be quickly solvable? | Would determine whether hard search problems have hidden shortcuts, with consequences for cryptography and optimization |
| Yang-Mills existence and mass gap | Does quantum Yang-Mills theory exist rigorously, and does it have a positive mass gap? | Would put the Standard Model's strong force on firm mathematical footing |
| Navier-Stokes existence and smoothness | Do smooth solutions to the 3D fluid equations always exist, or can they blow up in finite time? | Would settle whether turbulence can be tamed by the equations or whether singularities are real |
| Riemann hypothesis | Do all nontrivial zeros of the zeta function lie on the critical line? | Would sharpen the distribution of primes and thousands of downstream results |
| Birch and Swinnerton-Dyer conjecture | Does the rank of an elliptic curve equal the order of vanishing of its L-function? | Would connect arithmetic geometry to analytic number theory |
| Hodge conjecture | Are certain cohomology classes algebraic cycles? | Would bridge topology and algebraic geometry |
| Poincaré conjecture | Is every simply connected closed 3-manifold a 3-sphere? | Solved by Grigori Perelman, who declined the prize; the only resolved case |
The pattern is visible. Six of the seven remain open. One, the Poincaré conjecture, was resolved by Perelman in work circulated in 2002 and 2003. The Clay Institute awarded the prize in 2010, and Perelman declined it. That single resolved case is a useful calibration: even a solved Millennium problem took roughly a century after its statement and required entirely new mathematics.
The mechanism that makes these problems hard is not computational brute force. It is that each requires a proof that holds for all cases, not a verification that holds for many. A computer can check millions of examples of the Riemann hypothesis and find no counterexample. That checks nothing about the infinite tail. This distinction is exactly where claims about artificial intelligence solving such problems tend to break down.
Quantified context: the prize, the age, and the scale
The numbers behind the list are modest on their face and enormous underneath. The Clay Institute announced the problems on May 24, 2000, at the Collège de France in Paris. The prize is USD 1 million per problem. The problems themselves are older: the Riemann hypothesis dates to 1859, the Poincaré conjecture to 1904, and Navier-Stokes smoothness questions to work by Leray in the 1930s.
The time scale matters for judging any claim. Perelman's proof of Poincaré came about 98 years after the conjecture was posed. Andrew Wiles proved Fermat's Last Theorem in 1994, roughly 358 years after it was stated, and that problem was not on the Millennium list. The lesson is that major proofs are usually decades-long community efforts, not single announcements.
The Navier-Stokes problem itself has two branches. The 2D version is solved: smooth solutions exist for all time. The 3D version is open. That asymmetry is the counterintuitive detail many summaries miss. The same equations behave qualitatively differently in two and three dimensions, and the three-dimensional case is where turbulence, energy cascades, and possible finite-time blow-up live. A claim that "AI solved Navier-Stokes" without specifying which dimension and which regularity class is describing something other than the Millennium problem.
Timeline: how the list and its claims evolved
| Year | Event |
|---|---|
| 1859 | Riemann states his hypothesis on zeta zeros |
| 1904 | Poincaré poses his conjecture on 3-manifolds |
| 1930s | Leray develops weak solutions for Navier-Stokes, opening the regularity question |
| 2000 | Clay Mathematics Institute announces the seven Millennium Prize Problems |
| 2002–2003 | Perelman circulates proofs resolving the Poincaré conjecture |
| 2010 | Clay awards the Poincaré prize; Perelman declines |
| 2020s | Machine-learning systems begin assisting with proof search and conjecture generation, without resolving a Millennium problem |
The timeline is the antidote to hype. Progress in this domain is measured in decades and in structural breakthroughs, not in product announcements. Recent AI systems have made real contributions to mathematics, including helping discover patterns and assisting with formal verification. None has produced a peer-reviewed, accepted resolution of a Millennium problem. The gap between "assisted with a lemma" and "solved Navier-Stokes" is the entire distance between a calculator and a proof.
Edge cases: what counts as a solution
The boundary of a valid solution is stricter than most people assume. A proof must be accepted by the mathematical community, typically after peer review and often after years of checking. It must address the exact statement, not a variant. It must handle all cases the statement covers. And it must survive scrutiny by specialists who have spent careers on the problem. The bullets below are the red flags to watch when a headline claims a breakthrough.
The claim addresses a special case and generalizes without proof. The claim relies on numerical evidence or machine output that cannot be independently verified. The claim has not been peer reviewed and is announced through press release or preprint alone. The claim conflates a related solved problem, such as the 2D Navier-Stokes case, with the open 3D case. The claim uses nonstandard definitions that shift the problem's meaning. The claim has no independent expert endorsement beyond the authors.
These flags are not cynicism. They are the normal quality control of a field where a single gap in a proof can invalidate years of work. The Poincaré case is instructive: Perelman's proof took years to verify, and the verification itself was a major undertaking. A machine-generated proof would face the same scrutiny, and arguably more, because the reasoning must be checkable by humans or by trusted formal systems.
Misconceptions: the AI-solves-math narrative
The most common misconception is that a language model producing a plausible proof sketch equals a solved problem. It does not. Mathematical proof is a social and logical process, and plausibility is the starting point, not the finish line. The second misconception is that the seven problems are equally hard or equally close. They are not. Some, like the Poincaré conjecture, now have a resolution. Others, like the Riemann hypothesis, have resisted more than 160 years of attack.
A third misconception is that the prize is the point. The USD 1 million is a rounding error next to the consequences. A proof of P versus NP would reshape cryptography overnight: if P equals NP, many current encryption schemes become breakable in principle. If P does not equal NP, that negative result would still transform complexity theory. Either answer is a structural earthquake, and that is why the problems attract the effort.
The fourth misconception is that AI is irrelevant here. It is not. Machine learning already helps with pattern discovery, conjecture generation, and formal proof checking. The honest framing is that AI is a new tool in a long tradition of tools, from symbolic computation to automated theorem provers. Tools accelerate mathematicians. They have not replaced the need for a proof that the community accepts.
Practical actions: how to evaluate a breakthrough claim
If you encounter a claim that a Millennium problem has been solved, including the Navier-Stokes problem, apply a short checklist before believing or sharing it. The steps below are concrete and ordered.
Identify the exact statement being claimed, including dimension, regularity class, and definitions. Check whether the work has appeared in a peer-reviewed venue or has clear expert endorsement. Look for independent verification by specialists, not just the authors' summary. Distinguish assistance from resolution: a tool that helps find a lemma is not a proof of the whole problem. Treat press-release language as a signal to slow down, not speed up. Check the Clay Mathematics Institute's official status for the problem.
The threshold for belief should be high. A valid solution needs review by people who understand the field deeply, and that review takes time. The same standard applies whether the author is a human, a machine, or a collaboration. The question is never "did a system produce output that looks like a proof?" The question is "has the mathematical community accepted a proof?"
Limitations: what this explainer can and cannot settle
This article explains the list, the mechanism behind each problem, and the standards for a valid solution. It cannot settle any of the seven, and it does not adjudicate specific claims beyond the general flags above. The state of any individual problem should be checked against the Clay Institute's official records, which are the authoritative source for prize status. Claims about AI solving specific problems should be evaluated against peer-reviewed literature, not against headlines.
One limitation is worth stating plainly: the field moves, and a preprint can appear at any time. This explainer describes stable structure, not a live scoreboard. If a genuine solution is announced, the verification process will take months or years, and the community response, not the announcement, will be the signal. The Navier-Stokes problem in particular has a long history of claimed proofs that did not survive review. That history is the reason for caution, not a reason to dismiss the possibility.
The deeper point is that the Millennium Prize Problems are a map of mathematics' unresolved load-bearing questions. They are known collectively as the Millennium Prize Problems because the Clay Institute framed them as the field's defining challenges for the new century. Whether AI becomes part of how one of them is solved is an open question. Whether a claim of solution is credible is a question with a clear standard: proof, review, and acceptance.