Will Anthropic announce Claude has solved Navier-Stokes before July 2027?
Latest forecast
Generated Nov 20, 2026, 11:51 PM · High effort
Write-up
Executive Summary
Probability that the question resolves YES: 65.6%. Anthropic is more likely than not to make a qualifying announcement by June 30, 2027.
The decisive development is that a concrete Claude-assisted claimed solution already exists. On November 20, 2026, Levent Alpöge and Tristan Buckmaster released work claiming a breakdown of the unforced, three-dimensional Navier–Stokes equations—the kind of counterexample that could resolve the full Clay problem.
The research did not verify a qualifying official Anthropic announcement. Two hurdles remain: completing enough technical verification to support a definitive claim, and securing Anthropic’s adoption of that claim. This forecast concerns the announcement, not whether the mathematics is ultimately accepted.
Major update
A newly released Claude-assisted claimed solution has replaced a promising research route with a concrete candidate for the full Navier–Stokes problem. The chance of an official Anthropic solution announcement by June 30, 2027 rises from about 25% to 66%—roughly one in four to two in three—making an announcement the more likely outcome, though none was verified in the newer report.
Key drivers
A claimed solution now directly targets the question
The November 20 paper and accompanying repository claim finite-time blow-up: an initially smooth fluid flow develops a singularity within a finite time. The claimed construction is unforced, meaning it has no external force.
An appropriate unforced counterexample would resolve the full Millennium Problem under Clay’s official formulation. It is not automatically an excluded “special case”: one valid counterexample can disprove a universal existence claim. This makes the release much more consequential than a simulation, partial theorem, or extrapolation from mathematics benchmarks.
Claude’s involvement is explicitly documented
The paper’s AI statement says: “We used Anthropic models (Claude) throughout this work, and no others.” It describes human guidance and curated mathematical inputs, and acknowledges Anthropic’s support for the collaboration.
That supplies a strong basis for crediting Claude as a substantial contributor. It does not establish that Anthropic has endorsed the theorem or approved a company announcement.
The public verification package is materially incomplete
The release documentation says the decisive numerical certificate—the data intended to establish key proof conditions—and its checking computation are not included. A patch changing mathematical bounds and conditions had also not been rebuilt when released.
The distinction matters because much of the Lean formalization, a proof encoded for software checking, establishes a conditional chain: if suitable data satisfy specified conditions, breakdown follows. That alone does not establish that such data exist and satisfy the final conditions.
Terence Tao’s assessment recognized the general computer-assisted strategy but said further work was needed for a professionally written proof. This supports treating the candidate seriously, not treating it as independently verified.
Anthropic has both an announcement pathway and reasons for caution
Anthropic’s communications distinguish limited achievements from complete solutions. Its August 10 Riemann-zeta post explicitly said Claude had not solved the Riemann hypothesis. Its September 4 Fermat post distinguished formalizing known mathematics from discovering a new result. Conversely, its November 17 black-hole physics post used affirmative “solves” language after describing formal and human checks.
Author preferences create an additional barrier. Buckmaster’s statement characterized the earlier collaboration as personal, and Alpöge’s September statement expressed an intention to refuse marketing around the result then being pursued. But Buckmaster also supported loud corporate acknowledgment of a genuine breakthrough if its history remained intact. These statements justify caution about corporate adoption, not an assumption that respectful joint credit is permanently prohibited. Same-day corporate silence receives little weight by itself.
The background has changed since September
On September 8, 2026, OpenAI officially claimed a Navier–Stokes solution through the permitted breakdown alternatives with smooth external forcing. Clay’s September 11 statement described the problem as apparently settled while emphasizing an unhurried evaluation process.
Neither event resolves this question. They do establish a relevant precedent for a company making a full-solution claim before institutional certification. The criteria do not require Anthropic to announce first.
Scenarios
The reasoning separates the state of the mathematics from the decision to make an official claim. “Sufficiently substantiated” means enough auditable evidence and specialist support for a definitive announcement—not final Clay acceptance.
| Scenario | Why it carries this weight | Route to resolution |
|---|---|---|
| The current research becomes sufficiently substantiated | This receives the greatest weight: there is a full-target manuscript, relevant mathematical expertise, a recognizable proof strategy, and substantial formal work. Missing verification materials prevent near-certainty. | The clearest route to YES is publication and checking of the remaining artifacts, followed by an affirmative Anthropic statement with joint credit. Author preferences or cautious wording could still produce NO. |
| The claim remains credible but verification is incomplete or seriously disputed | This is a meaningful secondary scenario because the missing certificate, unrebuilt patch, and correspondence between the written and formal arguments require substantive scrutiny. | NO is favored within this scenario if Anthropic sticks to “candidate” or “under evaluation” language. A definitive company claim before disputes are settled could nevertheless produce YES. |
| The central construction fails | This receives the least weight, but remains plausible because the decisive public numerical evidence is absent. | Ordinarily this leads to NO. A premature qualifying announcement, or a replacement Claude-assisted construction before the deadline, could still lead to YES. |
These are judgment-based scenarios, not measured historical frequencies. Their weighting produces the final forecast.
Implications
- Prepare for a communications milestone without assuming a certified theorem. A qualifying announcement could arrive before broad mathematical agreement. Subsequent rejection or retraction would not reverse a YES resolution.
- Keep technical reliance separate from announcement monitoring. Decisions requiring correct mathematics should depend on the certificate, reproducible checking, and specialist scrutiny—not merely the corporate headline.
- Do not use the Clay prize timetable as the trigger. An official Anthropic statement can settle this question well before prize recognition. Conversely, an author-controlled paper or personal employee post cannot settle it by itself.
Signs to watch
The recorded research established no dated schedule for releasing the certificate, completing independent verification, or making an Anthropic announcement. The firm future date is the June 30, 2027 deadline. The following developments could occur without advance notice.
| Development and timing | Effect on the forecast |
|---|---|
| Certificate and checking computation are released, at any time before the deadline | Successful reproduction against the final proof conditions would make YES more likely. Failure to meet those conditions, or continuing inability to reproduce the result, would make it less likely. |
| The patched formal package is rebuilt and independently reviewed | A successful build with a clear match to the intended theorem would make YES more likely. A substantive mismatch or unrepairable gap would make it less likely. |
| Specialists publish detailed assessments | Confirmation of the central construction would make YES more likely; a confirmed fundamental defect would make it less likely. Corrections to exposition alone should not be mistaken for a refutation. |
| Authors or Anthropic clarify permission and framing | Support for an official joint-credit announcement would make YES more likely. An explicit refusal of company promotion, or a decision to avoid definitive solved-language, would make it less likely. |
| Anthropic publishes an official statement | An unambiguous full-solution claim with substantial Claude credit would settle YES, even outside a dedicated launch. A statement confined to a candidate proof would leave the question open. |
| June 30, 2027 ends without a qualifying statement | The question resolves NO, irrespective of the paper’s eventual fate. |
Reproducible completion of the verification package, a confirmed central defect, or a clear institutional decision would settle a major uncertainty and make the forecast worth rerunning.
Research and methodology
The forecast combines direct documentary review with a scenario calculation that treats technical substantiation and corporate adoption as separate hurdles. The main sources were:
- The Alpöge–Buckmaster manuscript and repository: the central evidence for the claimed theorem, Claude attribution, Anthropic support, and missing verification materials. The manuscript text was extracted and reviewed alongside the repository documentation.
- The formalization documentation: essential for distinguishing checked conditional implications from the still-unpublished numerical evidence needed to complete the argument.
- Tao’s November 20 assessment: an external assessment of the strategy and the work still required, rather than a complete validation.
- Clay’s official problem formulation and September statement: used to check mathematical scope and distinguish a claimed solution from institutional acceptance.
- Anthropic’s Riemann-zeta, Fermat, and black-hole physics publications: selected precedents for announcement wording and verification standards. They are not an empirical sample from which an announcement rate can be calculated.
- Buckmaster’s statement, Alpöge’s September statement, and searches of Anthropic’s Research and News channels: evidence about attribution, marketing preferences, and whether corporate adoption had already occurred.
Limitations
The decisive certificate and checking computation were unavailable for public reproduction. The research did not independently verify the mathematical argument or build the formal package, and could not establish that the intended data satisfy every condition after the patch.
The sources also established neither a promised Anthropic announcement nor a binding prohibition on one. Searches of official channels were bounded, so the absence of a verified qualifying statement is not proof that every possible company-controlled communication was captured.
Forecast history
Hover a data point to trace its series, or click to view the forecast generated at that time.
Signed forecast receipt
Signed Nov 20, 2026, 11:51 PM with ed25519 key preseen-prod-ed25519-20260523 and externally timestamped Nov 20, 2026, 11:51 PM.
sha256:53bc2f2ff77c4a...23c91af923Want forecasts like this?
Preseen researches your questions and answers them with calibrated, source-backed forecasts.