OpenAI announced that internal AI models had produced three significant research outputs spanning mathematics, computer science, and physics. The company described the work as part of an effort to collaborate with academic communities on verification and interpretation, releasing materials in different ways for each result and contributing funding toward related workshops.
The computer science claim has drawn particular attention because it challenges a longstanding conjecture used as a foundation for many conditional hardness results.
The Exponential Time Hypothesis (ETH) is a conjecture in computational complexity theory about the difficulty of Boolean satisfiability problems, specifically k-SAT.
It is closely connected to questions around P versus NP and has been widely treated as true for decades.
Researchers have relied on it to derive hardness results across graph theory, computational biology, databases, computational geometry, and other areas. OpenAI stated that one of its systems generated a natural-language disproof on October 16, 2026, which was then formalized in the Lean proof assistant the following day.
The company described the result as showing a deterministic algorithm that counts satisfying assignments of k-CNF formulas in subexponential time while using only polynomial space, a bound that would refute ETH along with related hypotheses such as Strong ETH.
OpenAI has published both the AI-produced manuscript and the Lean formalization on its public GitHub repository.
A group led by Russell Impagliazzo and Ramamohan Paturi of UC San Diego—who, with Francis Zane, originally proposed ETH—and Ryan Williams of MIT is organizing an independent examination.
These researchers have emphasized that their involvement does not constitute certification and that a claim of this scale requires careful scrutiny by the broader community.
OpenAI has framed the potential impact as reopening questions about how much faster certain computational tasks might become and as a guide for searching for improved algorithms, while stressing that practical gains would require substantial further work.
The same announcement included two other results.
In mathematics, models produced a claimed proof of the local smoothing conjecture in every dimension, based on a novel auxiliary partial differential equation; this would also resolve the Kakeya and Stein restriction conjectures.
Materials for that result were released publicly for independent review.
In physics, the systems generated claimed existence proofs for quantum chromodynamics and pure Yang-Mills theory in four dimensions, addressing part of a Clay Mathematics Institute Millennium Prize problem.
Those physics materials are being handled through an external group that has deferred public release of the AI artifacts while preparing its own write-ups.
None of the three claims has yet undergone full independent verification, and OpenAI has presented them as AI-generated outputs for expert assessment rather than settled results.
The company has stated it seeks no academic credit and has provided support for community discussion of the broader questions raised by AI-produced research, including verification, access, and credit.
It is important to note that the Exponential Time Hypothesis is unrelated to the cryptocurrency Ethereum, which shares the same acronym.
OpenAI’s announcement contains no claims about Ethereum, Bitcoin, smart-contract platforms, or the security of existing decentralized protocols.
While advances in complexity theory can, in principle, affect the hardness assumptions underlying various cryptographic systems over the long term, the current result is a theoretical claim about k-SAT counting algorithms whose practical implications—if the proof holds—remain to be determined after verification. Early-stage claims of this type require cautious interpretation until experts complete their reviews.
