🔍 Read the full analysis: The Exponential Time Hypothesis And AI: What OpenAI’s Result Does—and Doesn’t—Show on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI says an unreleased AI model produced a proof refuting the Exponential Time Hypothesis, a prominent assumption about the difficulty of solving 3-SAT. Lean has checked a formalized statement, but the formalization covers a slightly weaker runtime bound than the preprint, and independent expert review has not yet been completed. The result does not solve P versus NP or show that practical computing will become faster.
OpenAI said on October 26 that an unreleased internal AI model had produced a proof refuting the Exponential Time Hypothesis (ETH), a longstanding assumption about the difficulty of solving certain logic problems. A Lean formalization has been published and machine-checked, but independent expert review is still pending, and the formal statement checks a slightly weaker runtime bound than the preprint claims.
The preprint claims a deterministic algorithm that exactly counts solutions to any fixed-width k-SAT formula in subexponential time and polynomial memory. For 3-SAT, that would contradict ETH, which holds that solving such instances requires exponential time. The paper also says the result refutes the stronger Strong Exponential Time Hypothesis and three related hypotheses.
OpenAI says its model found the disproof on October 16, and that the Lean formalization followed the next day. The company released the preprint, formal statement and Lean code publicly at the request of the outside group handling the computer-science result. The source article reports counting 3,977 files and about 710,000 lines of Lean code; those are its own counts, not figures independently confirmed here.
Lean’s check is not the same as expert review. It verifies that the formalized proof follows from its stated assumptions, but the published formalization covers a slightly weaker runtime bound than the sharpest bound in the paper. The source account says a group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams will examine the work. Impagliazzo and Paturi were among ETH’s original authors.
THREE RESULTS, THREE RELEASE STRATEGIES
What OpenAI released, what Lean checked, who reviews it
Dates and the physics status are OpenAI's own account. Lean-checked means a machine verified the proof, not that experts have accepted it.
Local smoothing, so Kakeya too
- AI output
- 18 Oct 2026
- Lean
- 19 Oct 2026 · strict ranges
- Public
- AI write-up + Lean proof
- Review
- No group: researchers organize themselves
The Exponential Time Hypothesis is false
- AI output
- 16 Oct 2026
- Lean
- 17 Oct 2026 · weaker bound
- Public
- 67-page preprint + Lean proof
- Review
- Impagliazzo, Paturi (UC San Diego), Williams (MIT)
QCD exists; pure Yang–Mills as sub-result
- AI output
- 9 Sep (Yang–Mills), 18 Sep (QCD)
- Lean
- Yang–Mills existence, 16 Oct (OpenAI)
- Public
- No: the group holds the AI materials back
- Review
- Gross, Douglas, Jaffe, Klebanov, Nekrasov
Before this release
The repository the results sit in
- 6 Oct: more than 700 AI-generated manuscripts published
- 7 Oct: 3 withdrawals after a sign error; 14 other manuscripts revised
- About 42% of top-line results formalized in Lean (repository’s own figure)
- OpenAI calls the manuscripts “raw research data”, not polished papers
Sources: OpenAI, 26 Oct 2026 · Lean note 378 · Lean note 079 · repository history. Status chips: our assessment.
Powered by Thorsten Meyer AI
What an ETH Refutation Would Change
If the preprint’s result survives scrutiny, it would overturn a widely used assumption in theoretical computer science. ETH underpins conditional lower bounds: researchers use it to argue that particular algorithms cannot run substantially faster than known approaches. A refutation would require those results to be reconsidered, or their assumptions to be made explicit.
That would not automatically produce useful speed-ups for businesses or consumers. The claimed algorithm is still superpolynomial, and OpenAI says practical improvements would require substantial further research. The source account also says the runtime’s improvement over exponential search grows very slowly with input size. For now, the near-term significance is primarily theoretical and methodological: an AI-generated mathematical claim has been paired with a machine-checked formalization and a planned external review.
The episode also puts attention on how AI research claims are validated. A proof assistant can check formal steps in a statement encoded for it; it does not by itself establish that the encoding captures every important claim in a paper, or that the result is useful. The gap between the preprint’s sharpest bound and the Lean-checked bound is one reason readers should distinguish the two.
Popular right now
Dyson Car+Boat Cordless Handheld Vacuum
Ninja Luxe Café Mini Plus Espresso and Drip Coffee Machine with AutoFroth, Stainless Steel, ES351
iRobot Roomba Max 715 Pet Robot Vacuum with AutoEmpty Dock
As an affiliate, we earn on qualifying purchases.
ETH, SAT and the OpenAI Release
k-SAT asks whether variables can be assigned true or false so that every clause in a formula is satisfied; each clause contains at most k variables. A formula with n variables has 2ⁿ possible assignments, though algorithms can do better than checking them all. ETH, proposed by Impagliazzo, Paturi and Francis Zane in 2001, says that for 3-SAT the exponential cost cannot be eliminated: the problem requires 2^(c·n) time for some fixed positive c.
ETH is not the same as P versus NP. P versus NP asks whether every problem whose proposed solutions can be checked efficiently can also be solved efficiently. The Clay Mathematics Institute lists that question as unsolved. Refuting ETH would challenge a stronger assumption about computational hardness, but would not prove that P equals NP or settle the P-versus-NP problem.
The ETH claim appeared alongside two other results in an OpenAI announcement dated October 26. The company said one concerns a proof of Sogge’s local smoothing conjecture and related conjectures in harmonic analysis; another concerns existence results in quantum chromodynamics and a grand unified completion of the Standard Model. According to OpenAI, outside groups are guiding how those manuscripts are handled. The source account also notes that OpenAI published more than 700 AI-generated manuscripts on October 6 and later withdrew three after a sign error, revising 14 others. That history adds a practical reason to examine verification and correction processes carefully.
The Proof Still Needs Expert Review
The central open question is whether independent specialists will validate the paper’s reasoning and its claimed implications. Lean has checked the formalized statement, but that statement is slightly weaker than the sharpest runtime bound in the preprint. The source material says an external group led by Impagliazzo, Paturi and Williams will examine the work; it does not provide the group’s findings or a completed review.
It is also not yet clear how the formalization relates in detail to every step and assumption in the full paper, or whether the claimed refutations of the stronger hypothesis and the three related hypotheses will withstand scrutiny. The source account does not specify a timetable for the review. Until reviewers report back, the result should be described as OpenAI’s announced claim, supported by a public preprint and a machine-checked formal statement, rather than as a conclusion endorsed by the field.
Reviewers Will Test the Claim
The next milestone is the outside group’s examination of the computer-science result. That review can clarify whether the proof establishes the paper’s full claims, how the difference between the paper’s runtime bound and the formalized bound should be understood, and whether any corrections are needed. OpenAI says it has agreed to fund a conference on the result, which could provide a forum for further technical discussion.
Readers should look for the reviewers’ account, any revised preprint or Lean code, and independent responses from complexity theorists. Until those appear, the formal check is meaningful but not a substitute for peer scrutiny, and the claimed result should not be read as a practical breakthrough or a resolution of P versus NP.
Key Questions
What does OpenAI say its AI proved?
OpenAI says an unreleased model produced a proof refuting the Exponential Time Hypothesis. The preprint claims a deterministic, subexponential-time algorithm for exactly counting solutions to fixed-width k-SAT formulas; for 3-SAT, that would contradict ETH.
Has the result been independently verified?
Not according to the supplied account. A Lean formalization has been machine-checked, and an outside group is expected to examine the result. The independent review has not yet been reported as complete.
Does this mean P equals NP?
No. ETH is a stronger assumption about the time needed for problems such as 3-SAT. Refuting it would not settle P versus NP, which remains an open problem.
Will computers solve hard problems faster now?
There is no evidence in the supplied material of an immediate practical speed-up. The claimed algorithm remains slower than polynomial time, and OpenAI says practical improvements would require further research.
What is the difference between the preprint and Lean check?
The preprint presents the authors’ full mathematical argument and its sharpest runtime bound. The published Lean formalization checks a slightly weaker bound, according to the source account, so the two should not be treated as identical claims.
Source: ThorstenMeyerAI.com
