The six open Clay Millennium Prize Problems, stated precisely, with the honest state of each and the computable ground Neuruh is mounting an attempt on. The atlas is build one of the program. The lab comes next.
On May 24, 2000, at the Millennium Meeting at the College de France in Paris, the Clay Mathematics Institute named seven problems and pledged $1,000,000 for the first correct solution to each. John Tate and Michael Atiyah announced them. The seven span number theory, algebraic geometry, topology, partial differential equations, theoretical computer science, and mathematical physics. The Institute modeled the prize on David Hilbert's 23 problems of 1900, which drove a century of mathematics.
One of the seven is solved. Grigori Perelman proved the Poincare conjecture, completing Richard Hamilton's program, in papers released in 2002 and 2003. The Clay Institute awarded him the prize in 2010. He declined it, because it was not also offered to Hamilton, whose work the proof built on. He had already declined the Fields Medal in 2006. Six problems remain, and $6,000,000 of prize money is still unclaimed.
The Clay rule that matters most for this page: the Institute only considers a proposed solution after at least two years have passed since its publication. A claim made in 2026 cannot be an official solution in 2026, no matter how strong it looks. Verification is deliberately unhurried.
2026 is the year machine-assisted claims arrived in earnest. In September, OpenAI posted a claimed Navier-Stokes blowup with a Lean formalization. In October, OpenAI announced a claimed Yang-Mills existence proof, also formalized in Lean. In August, a report said an Anthropic research model raised the proven proportion of Riemann zeros on the critical line from about 41 percent to about 67 percent. A company called AIX Global claims its quantum system resolved all six problems in July 2026. None of these is an official solution. Each one is sorted where it belongs, in its own section and again in section 09.
| Problem | Posed | Status | 2026 claim |
|---|---|---|---|
| Birch and Swinnerton-Dyer | Early 1960s | OPEN | Formula proposal, Oct 7, 2026. Proposed, not accepted. |
| Hodge | 1950 | OPEN | No credible 2026 claim. |
| Navier-Stokes | 2000 (equations 1822) | CLAIMED | OpenAI blowup claim, Sep 8, 2026. Lean formalization. Awaiting independent validation. |
| P versus NP | 1971 | OPEN | Unreviewed preprint, Jan 2026. Not accepted. |
| Riemann Hypothesis | 1859 | OPEN | Proportion claim, Aug 2026, about 67 percent. Manuscript unreleased. Unverified. |
| Yang-Mills | 2000 (theory 1954) | DISPUTED | Existence claimed Oct 26, 2026. Assessed Nov 10, disputed in November. Mass gap not claimed. |
An elliptic curve is a cubic equation with a group law on its solutions. The rank counts how many independent rational points the curve has. The L-function is an analytic object built from counting the curve's points modulo every prime. The conjecture says the algebra and the analysis agree exactly: the number of independent rational points is read off from how the L-function vanishes at s = 1. The stronger form of the conjecture also predicts the leading coefficient of the L-function at s = 1, in terms of the regulator, the Tate-Shafarevich group, the real period, the Tamagawa numbers, and the torsion.
A complete bridge between the arithmetic of rational points and the analysis of L-functions. It would settle the oldest question about elliptic curves, which are the same objects that carry modern elliptic-curve cryptography, and it would complete the model case for the general Langlands program, which expects L-functions to control arithmetic everywhere.
October 7, 2026: OpenAI proposed a full BSD formula at low Selmer corank. Proposed, not independently accepted. It is a proposal, not a verification.
A variety is a geometric shape cut out by polynomial equations. Its topology has holes, and cohomology counts them. Some cohomology classes look like they could come from real geometric pieces sitting inside the shape. Those are the Hodge classes. The conjecture says every candidate is real: every Hodge class is built out of actual subvarieties. It is a claim that the topology of a variety is controlled by its geometry.
The unification of algebraic topology and algebraic geometry. A proof would say that the continuous, topological description of a shape and the discrete, equation-based description agree at every level, and it would give a machine for turning topological information into geometric objects.
No credible 2026 claim. Preprint servers carry unreviewed papers claiming proofs; none has been peer-reviewed or accepted.
The Navier-Stokes equations describe how fluids move: water, air, blood. They are Newton's second law applied to a fluid. The question is whether a perfectly smooth flow can break down. Can the velocity at some point grow without bound in finite time, so the equations stop making sense? Or does every smooth start stay smooth forever?
The mathematics of turbulence, weather, flight, and blood flow. A proof of global regularity would certify the equations that every simulation already trusts. A proof of blowup would show the equations themselves can fail, and would say exactly how a fluid breaks.
September 8, 2026: OpenAI posted a claimed solution. The claim is a finite-time singularity, a blowup, with a smooth applied force. The fluid starts at rest, the force is smooth, the total energy stays finite, and the velocity grows without bound in finite time. The claim establishes statements (C) and (D) of the official Clay formulation, the blowup alternatives. The writeup is 166 pages with a Lean formalization, produced by about 10,000 coordinating agents in about 88 hours, reaching the resolution on September 5. OpenAI says it will not claim the million-dollar prize.
A priority dispute followed. Tristan Buckmaster and Levent Alpoge announced related forced-Euler results on September 7, 2026, and the question of who did the mathematics first is unresolved. The Clay Institute only considers proposed solutions after at least two years from publication, so the clock has not started. The Institute has said the problem has apparently been settled, but its review is deliberately unhurried. As of November 20, 2026, no independent verification of the OpenAI claim has been established.
What the claim does not do: prove that smooth solutions exist for all time. It claims the opposite direction, blowup, under a smooth applied force. The Clay problem is stated as a choice between those two outcomes, so a verified blowup would settle it. Verification is the open step.
P is the set of problems a computer can solve quickly. NP is the set of problems whose solutions a computer can check quickly, given the answer. Every problem in P is in NP, because solving it is a way of checking it. The question is whether the reverse holds. Is finding an answer as easy as checking one? Every lock you can test a key in quickly: can you also find the key quickly?
Either answer remakes computing. If P = NP, modern cryptography collapses, because every code that is easy to verify and hard to break becomes easy to break, and optimization, search, and design become routine. If P is not NP, the fact the whole security economy already bets on is certified: verification is fundamentally cheaper than discovery. That gap is the foundation of every password, every signature, and every encrypted message.
Three formal barriers have been proved, and each one rules out a whole family of techniques. A proof must evade all three at once.
No accepted 2026 claim. A January 2026 arXiv preprint claims a technique called semi-relativization that evades all three barriers and separates P from NP. It is unreviewed and not accepted.
The zeta function encodes the primes. Its zeros, the points where it equals zero, control how the primes are distributed. Riemann conjectured in 1859 that every interesting zero sits exactly on one vertical line, where the real part equals one half. If true, the primes are distributed as regularly as the analytic structure of zeta permits. If false, a single zero off the line breaks the pattern.
The sharpest possible error term in the prime number theorem, and hundreds of theorems currently proved on the assumption that RH holds would become unconditional. Prime generation, cryptography, and everything built on the distribution of primes rests on the same structure.
August 2026: a report says an unreleased Anthropic research model, about 60 coordinated subagents working for about a day and a half, produced a 35-page manuscript proving that at least 67.25 percent of non-trivial zeros lie on the critical line and are simple, and that at least 83.625 percent are distinct. Status: CLAIMED, UNVERIFIED. The manuscript is unreleased and no independent verification has been established as of November 20, 2026. The hypothesis itself is untouched either way.
Yang-Mills theory is the mathematics of the strong force, the force that holds atomic nuclei together. The problem asks for two things. First, build the quantum version of the theory rigorously, from the axioms up, in four-dimensional space. Second, prove it has a mass gap: the lightest particle it predicts has positive mass. The mass gap is why the strong force has a finite range and why protons have mass.
The first rigorous interacting quantum field theory in four dimensions. It would put the mathematical foundation of the Standard Model of particle physics on solid ground, and it would explain, from first principles, why the strong force behaves the way it does.
October 26, 2026: OpenAI announced an AI-generated existence proof for quantum chromodynamics, including pure Yang-Mills theory as a subproblem. The Lean formalization was submitted on October 28, 2026. The mass gap was explicitly not claimed.
November 10, 2026: an independent assessment, prepared by Michael R. Douglas, Arthur Jaffe, Ron Nissim, and Fred Rajasekaran and approved by the Yang-Mills group, with consultative review from David Gross, Igor Klebanov, and Nikita Nekrasov, judged that the submission proves a relativistic quantum field theory satisfying the Wightman axioms can be constructed as a scaling limit of four-dimensional Wilson lattice gauge theory for any simple compact gauge group. The same assessment says the submission does not prove uniqueness of the infinite-volume limit and does not prove a mass gap. Jaffe co-authored the official Clay problem statement with Edward Witten.
A November 2026 critique by Jenny Lorraine Nielsen argues the formalization does not establish existence in the Clay sense at all: the non-triviality test does not exclude generalized free fields, the contact terms are uncontrolled, and the identification of the constructed object as Yang-Mills theory, rather than some other lattice-originating theory, has not been verified by a human mathematician. Status: existence contested, mass gap open, Clay problem open.
Every claim from 2026, with its verdict as of November 20, 2026. The rule for the whole page: officially unsolved until Clay says otherwise. A Lean check proves that the encoded statement follows from its stated assumptions. It does not by itself prove the Clay statement. Independent human verification is the gate.
| Claim | Date | Verdict |
|---|---|---|
| OpenAI: Navier-Stokes blowup Finite-time singularity with smooth forcing, statements (C) and (D), 166-page writeup, Lean formalization, about 10,000 agents in about 88 hours. |
Sep 8, 2026 | CLAIMED Awaiting independent validation. Priority dispute open. Clay two-year window not started. |
| OpenAI: Yang-Mills existence Existence of quantum Yang-Mills as a Wightman QFT, Lean formalization submitted Oct 28, 2026. Mass gap not claimed. |
Oct 26, 2026 | DISPUTED Independent assessment Nov 10 judged the Wightman construction proved. A November critique says it fails the Clay criteria. Mass gap open. |
| Anthropic: Riemann proportion, about 67 percent About 60 subagents, about a day and a half, 35-page manuscript, at least 67.25 percent of zeros on the line and simple. |
Aug 2026 | UNVERIFIED Manuscript unreleased. No independent verification established. |
| OpenAI: BSD formula at low Selmer corank Full BSD formula proposed for curves of low Selmer corank. |
Oct 7, 2026 | UNVERIFIED Proposed, not independently accepted. |
| AIX Global: all six solved by quantum computation Claims formal resolutions of all six via Seed IQ governed fault-tolerant quantum computation, released Aug 30, 2026. |
Jul to Aug 2026 | UNVERIFIED Self-published on Medium and Zenodo. No independent confirmation. Not presented as solved anywhere on this page. |
| Preprint: P versus NP by semi-relativization Claims a technique evading all three barriers and separating P from NP. |
Jan 2026 | UNVERIFIED Unreviewed preprint. Not accepted. |
This atlas is build one. It does what an atlas does: it says what is open, what is known, what is claimed, and where the computable ground is. Build two is the Millennium Lab, and it is a real program, not a claim.
The lab points the sovereign runtime at the computable frontiers this atlas names. The runtime already exists and is proven: 117 pytest tests passed, 61 receipts in an end-to-end run, hash chain valid after restart, gates that dispatch once and resume on a signed founder token. The lab adds Millennium computations to that runtime.
| Problem | Computable frontier | Lab first move |
|---|---|---|
| Riemann | Verify zeta zeros to a height | A verified zero count with hash-chained receipts, re-runnable by anyone. |
| Birch and Swinnerton-Dyer | Compute ranks and L-values for a curve family | Rank and L-value computation for a chosen family, receipted per curve. |
| Navier-Stokes | Numerical blowup search on a small grid | A small-grid blowup search with every run receipted, and a reading of the OpenAI formalization against the Clay criteria. |
| P versus NP | Formalize recipe search as a complexity object | Recipe search written as a complexity object, audited against the three barriers. |
| Yang-Mills | Wilson lattice computation | A small Wilson lattice computation with receipts, plus a gap audit of the OpenAI formalization. |
| Hodge | Machine-checkable statements in Lean | Formalization of the cycle class map statement, starting with the cases already proved. |
No solution claims. A real program with receipts. The lab page comes next, and it will carry the first computation, not the first promise.