§ Claim under review · Fact
"OpenAI announced that an internal AI system, more capable than GPT-6 Astra, produced a formalized and verified proof solving the Navier-Stokes existence and smoothness problem, though mathematicians Tristan Buckmaster and Levent Alpöge allege OpenAI may have accessed their unpublished research to achieve this."
Verdict
Partially accurate but misleading
Confidence
HighSummary
OpenAI really did announce on 8 September 2026 that an internal AI system, which it described as significantly more capable than GPT-6 Astra, produced a proof and a Lean formalization of finite-time blowup for the 3D Navier-Stokes equations. The important detail the post leaves out is scope: the theorem covers the version of the equations driven by a smooth external force, and the unforced version that most people have in mind is not addressed. The proof is machine-checked in Lean, but as of late September 2026 it had not been independently refereed, and the Clay Mathematics Institute had not accepted it, describing its own review as unhurried. The post also overstates the dispute. The public statement came from Tristan Buckmaster, made partly on behalf of Levent Alpöge, and in it he writes that he does not know whether their data was used and that he is not accusing anyone of anything. OpenAI said no specific user data was accessed, while also saying it cannot rule out that de-identified data from the pair's product usage helped improve its models, and the post reports that part correctly. What remains open is whether the proof survives independent review and whether the mathematical community treats the forced-case result as resolving the Millennium problem.
The readings
key figures from the evidenceagents used by OpenAI system to produce the proof
duration of AI system's work producing the proof
Why this verdict
Evidence
OpenAI published a page on 8 September 2026 stating that a multi-agent system running on an unreleased internal model produced an analytical proof and a Lean formalization of finite-time singularity formation for the 3D incompressible Navier-Stokes equations, and that this "resolves the Navier–Stokes Millennium Prize problem by establishing statement 'C' (and also 'D')" of the official Clay formulation. Company and press accounts describe roughly 10,000 concurrent agents, about 88 hours of work, and a further period of Lean formalization and checking.
The proved theorem, as restated in two independent arXiv preprints, constructs for every positive viscosity a smooth, compactly supported external force and a smooth solution starting from rest whose velocity becomes unbounded in finite time. One of those preprints states plainly that this addresses Clay alternatives C and D, breakdown with a smooth force, and that "the unforced alternatives are untouched", and records that as of 12 September 2026 there was no refereed independent verification.
On the dispute: Buckmaster's own statement, posted hours before the OpenAI announcement, describes his collaboration with Alpöge, describes being told by OpenAI that an internal model had proved finite-time blowup for the forced equations, and describes a contentious exchange over authorship. In that statement he writes that he does not know whether their data was used and "I am not accusing anyone of anything." OpenAI's public response says its researchers and agents did not see the pair's work until public release and that "no specific user data was accessed in order to solve this problem", while adding that "we cannot rule out that de-identified data derived from their usage of our products helped improve our models."
On official status, secondary reporting and reference pages indicate the Clay Mathematics Institute has not accepted the result, has described the problem as apparently settled while saying its evaluation will be unhurried, and continues to list the problem as active pending peer-reviewed publication and community vetting.
Findings
✓ What's accurate 6
- OpenAI did publicly announce, on 8 September 2026, that an internal AI system produced a proof of finite-time singularity formation for the 3D Navier-Stokes equations, accompanied by a Lean formalization.
- OpenAI described the system behind the proof as an unreleased internal model significantly more capable than GPT-6 Astra, and a formalization step involving GPT-6 Astra is described in coverage of the announcement.
- OpenAI itself framed the result as resolving the Millennium Prize problem by establishing statements C and D of the official Clay formulation.
- A Lean formalization was released publicly, and at least one preprint reviewing the artifact describes the formal development as recording no unproved steps and permitting only standard library axioms.
- A dispute over priority and data access did follow, involving Buckmaster at NYU and Alpöge at Anthropic, and it was reported by named outlets including Axios, CNBC, TechCrunch and Quanta.
- OpenAI did deny that its researchers or agents saw the pair's work before public release, and did state that it cannot rule out that de-identified data derived from their product usage helped improve its models. The post's caption reports both of those points correctly.
≈ What's misleading 5
- The claim says the proof solves "the Navier-Stokes existence and smoothness problem" with no qualification. The theorem, as restated in independent preprints, establishes blowup for the equations driven by a smooth compactly supported external force, which addresses Clay alternatives C and D; the unforced case that most readers associate with the problem is not addressed. A reader is likely to conclude the whole famous question is closed.
- The claim calls the proof "verified" without saying what kind of verification. What exists is a machine-checked Lean artifact. Independent refereed verification was not complete as of mid-September 2026, and reporting through 24 September 2026 described mathematicians as still working through the argument.
- The claim presents a settled outcome in the present tense. The Clay Mathematics Institute had not accepted the result, and reference pages still listed the problem as active or unverified as of late September 2026, so the status stated is ahead of the official record.
- The claim says Buckmaster and Alpöge "allege" that OpenAI may have accessed their unpublished research. The record is a statement authored by Buckmaster, made partly on behalf of Alpöge, not a joint allegation by two named accusers.
- The claim presents the allegation more strongly than its source. Buckmaster's statement says he does not know whether their data was used and states that he is not accusing anyone of anything; what he describes is a sequence of events, a suspicion about timing and approach, and a contentious authorship discussion.
? What's uncertain 4
- Whether the OpenAI result will be accepted by the Clay Mathematics Institute or by the mathematical community, and on what timeframe. Clay's stated posture in secondary reporting is that its review will be unhurried and that recognition follows peer-reviewed publication.
- Whether the analytical manuscript, as distinct from the Lean artifact, holds up under independent reading. Reporting in late September 2026 described mathematicians as still checking it.
- Whether any data derived from Buckmaster's or Alpöge's use of OpenAI products influenced the model. OpenAI says specific user data was not accessed but says it cannot rule out an indirect training benefit from de-identified data, and no independent audit of that question was found.
- Whether OpenAI's forced-case theorem is generally accepted as satisfying the Millennium problem. The Clay problem description does include breakdown alternatives that permit a smooth force, so part of the remaining disagreement is definitional rather than factual.
Sources
11 of 11 linked to recordsOpenAI, "On the Navier–Stokes Millennium Prize Problem" (company announcement page, writeup and Lean project linked)
Tristan Buckmaster, personal statement PDF hosted at NYU Courant
OpenAI official X statement responding to the allegations
arXiv preprint recasting the OpenAI construction (states the theorem's exact scope and verification status as of 12 Sep 2026)
arXiv preprint on regularity with analytic forcing, restating OpenAI Theorem 1.1
Axios, "OpenAI's historic math solution overshadowed by credit controversy"
CNBC, 9 Sep 2026 report
Quanta Magazine, 8 Sep 2026
TechCrunch, 8 Sep 2026
Wikipedia, "Navier–Stokes priority controversy" and "Navier–Stokes existence and smoothness" (current status summaries)
Tufts Daily, 24 Sep 2026, on continuing verification and Clay's posture