Key Notes
- OpenAI and Anthropic have published AI-generated arguments claiming to refute the Exponential Time Hypothesis, with formal Lean proofs available for examination.
- Anthropic explicitly describes its result as asymptotic, while OpenAI’s external review organizers caution that their involvement does not certify the claim.
- A valid ETH refutation would challenge an important assumption in complexity theory without automatically proving P equals NP or delivering practical speedups.
OpenAI and Anthropic have published AI-generated arguments claiming to refute the Exponential Time Hypothesis, a central assumption about the limits of computation. If the claims withstand scrutiny, researchers would have to revisit an important foundation of algorithmic hardness.
Both releases include formal proofs in Lean, giving specialists material they can check rather than only a company announcement. But formal verification and scientific acceptance address different questions. The experts organizing examination of OpenAI’s result have explicitly cautioned that their participation does not certify it.
What the Exponential Time Hypothesis Actually Says
ETH concerns 3-SAT, a problem that asks whether a collection of logical constraints can all be satisfied by assigning true or false values to variables. Its clauses contain at most three literals, such as a variable or its negation. The hypothesis says there is no algorithm that solves every such instance in subexponential time as the number of variables grows.
That is a claim about the hardest cases and long-run growth, rather than a claim that every individual puzzle is difficult. A particular formula can be easy, and practical solvers can perform well on many workloads, without contradicting ETH. Refuting it would require an algorithm with a sufficiently strong guarantee across all valid inputs.
Nor would the result automatically settle P versus NP. A subexponential algorithm can still be much slower than a polynomial-time one. Finding a way below the growth rate ETH rules out is a different achievement from showing that every efficiently checkable answer can also be found efficiently.
Claude’s Result Comes With an Explicit Practical Limit
Anthropic’s research folder describes a deterministic algorithm for counting satisfying assignments, a harder task than merely deciding whether any assignment exists. The company attributes the discovery to an internal research Claude model and says AI models also wrote the Lean formalization.
Its documentation is explicit that the result concerns asymptotic behavior and establishes nothing about solving formulas efficiently in practice. Large hidden constants or slowly improving bounds can separate a theoretical advance from a useful solver. Businesses should not read the announcement as an immediate shortcut for scheduling, optimization or other difficult computing tasks.
The verification notes also identify a human responsibility: checking that the formal statement accurately expresses the paper’s claim. The published process includes an independent proof-checking kernel alongside Lean. We inspected the documentation, but did not rebuild the full proof projects or independently validate either team’s theorem.
OpenAI’s own scope notes distinguish the bound established by its Lean formalization from sharper bounds in the accompanying manuscript. Both are presented as subexponential, but that difference illustrates why the exact theorem checked by software matters as much as the headline attached to it.
The Reviewers Are Inviting Scrutiny, Not Declaring Victory
OpenAI’s October 26 announcement names Russell Impagliazzo, Ramamohan Paturi and Ryan Williams as leaders of the ETH examination. Impagliazzo and Paturi were among the hypothesis’s original proposers. The company says its model found the argument on October 16 and formalization followed the next day.
In their open letter, the researchers separate checking this particular claim from the wider question of how computer science should handle AI-generated discoveries. They want an examination open to independent criticism, with explanations, gaps, corrections and verified steps shared across the community.
They also reject the idea that changing ETH’s status would erase the value of decades of research. Conditional conclusions might need reassessment, while the reductions and techniques connecting problems could remain useful. That is a more precise consequence than declaring hundreds of papers worthless.
OpenAI Also Claims Advances on Waves and Quantum Fields
The same release claims a proof of the local smoothing conjecture in every dimension, concerning how waves concentrate. OpenAI says it would also resolve the Kakeya and Stein restriction conjectures. A write-up and Lean proof are public.
A separate claim concerns rigorous existence of certain four-dimensional quantum field theories, including quantum chromodynamics. A group led by Nobel laureate David Gross will examine that work; its AI-produced artifacts are being withheld while the group prepares an account. OpenAI describes the Yang–Mills existence component as part of the Millennium Prize problem.
The qualification is important. The Clay Mathematics Institute’s problem statement requires both existence and a positive mass gap. An existence claim alone does not establish that the full prize problem has been solved.
The announcements follow OpenAI’s 722-manuscript release and the debate over researchers’ roles explored in our mathematics coverage. The next meaningful development will be what specialists can reconstruct, verify and explain from these arguments. Producing a proof and making its ideas usable are distinct contributions to science.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
