OpenAI says its AI disproved the exponential time hypothesis, solved a 1991 math conjecture, and advanced Yang-Mills theory. Here's what that means.
What did OpenAI actually claim? #
OpenAI says its AI models produced three new results across mathematics, computer science, and physics: a proof that disproves the exponential time hypothesis, a proof of the local smoothing conjecture and the related CAA (Kakeya) conjecture in every dimension, and a proof of existence for Yang-Mills theory, the mathematical backbone of the strong nuclear force. The announcement came from OpenAI researcher Dan Roberts, with president Greg Brockman amplifying it as a sign that AI is starting to help expand human understanding of the universe, not just write emails or make pictures.
TL;DR #
- OpenAI announced three claimed scientific discoveries in one batch: a computer science result about the exponential time hypothesis, a math result about wave behavior (the local smoothing conjecture), and a physics result tied to Yang-Mills theory.
- The exponential time hypothesis claim is the most startling one because it challenges a 27-year-old belief that certain hard problems (scheduling, routing, chip layout) have no fundamental shortcut around brute-force search.
- The math proof solved a problem open since 1991 by proving the local smoothing conjecture in every dimension at once, something human mathematicians had only managed dimension by dimension, with the 3D case cracked as recently as 2025.
- The physics result only solves part of a Millennium Prize problem : OpenAI claims a proof of existence for Yang-Mills theory, not the “mass gap” portion, so the $1 million prize is still unclaimed.
- None of this has been independently verified yet. The computer science and physics results are being reviewed by outside experts, including the original authors of the exponential time hypothesis and a group of physicists led by Nobel laureate David Gross.
- P versus NP is untouched. Brockman himself noted that P still probably doesn’t equal NP, so encryption and internet security are not suddenly at risk.
- The real-world payoff, if the claims hold, shows up quietly : faster chip design, better delivery routing, improved medical imaging math, not as a single dramatic headline moment.
Seven tools to build an app. Or just Remy. #
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
What is the exponential time hypothesis, and why does disproving it matter? #
The exponential time hypothesis (ETH) is a long-standing belief in theoretical computer science, formalized in 1999 by Russell Impagliazzo and Ramamohan Paturi. It says that for a certain class of hard problems, there’s no clever workaround to the brute-force explosion of possibilities. Think of a padlock: add one more digit to the combination, and the number of options doesn’t grow a little, it multiplies. Problems like vehicle routing, school timetabling, chip layout, and airport scheduling all have this same explosive structure at their hardest versions.
For 27 years, ETH was treated as a working assumption. It was never proven, but hundreds of other results were built on top of it, in the style of “if ETH is true, then this other problem must also be hard.” OpenAI claims its AI found a method that beats that explosion for the classic version of this problem. It’s still slow by everyday standards, but fundamentally faster than what the field believed was the limit. This is not the same as solving P versus NP, the most famous open problem in computer science. Proving P equals NP would mean every one of these hard problems has a genuinely fast solution. OpenAI’s claimed result doesn’t go that far, and Brockman explicitly said P still probably doesn’t equal NP. The proof was reportedly checked using Lean, a formal verification tool that checks mathematical proofs step by step, similar to a spell-checker for logical reasoning. OpenAI says it has brought in Ryan Williams, a leading figure in computational complexity, who is now working with Impagliazzo and Paturi, the original authors of the hypothesis, to scrutinize the claim.
How does the math proof relate to things like Wi-Fi and MRI scans? #
The second claimed result sits in harmonic analysis, the branch of math that describes how waves behave. Sound, light, radio signals, and Wi-Fi are all waves, and the mathematics of how waves spread, bunch up, and smooth out underpins technology most people use every day: file compression for streaming audio, MRI image reconstruction, photo compression on phones, and noise-canceling headphones.
The specific target was the local smoothing conjecture, proposed by mathematician Christopher Sogg in 1991, which asks how much waves naturally smooth out as they spread. It’s connected to the Kakeya (CAA) conjecture, a decades-old puzzle originally framed as a question about spinning a needle in the smallest possible space. Progress on these problems has historically been slow and dimension-specific: the two-dimensional Kakeya case wasn’t settled until around 2019, and the three-dimensional version was only cracked in 2025, celebrated at the time as one of the year’s biggest math results.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
OpenAI says its models proved the local smoothing conjecture in every dimension simultaneously, rather than one dimension at a time. According to the timeline shared by OpenAI, a first proof appeared on October 18, with a computer-checked version following the next day. OpenAI published the work on GitHub so mathematicians anywhere can inspect and verify it, which is the actual test that matters here, not the announcement itself.
What does the physics claim actually prove, and what’s still missing? #
The third result touches quantum chromodynamics (QCD), the theory describing the strong force that holds atomic nuclei together. QCD has worked remarkably well for about 50 years, matching experimental predictions with high accuracy. But nobody had mathematically proven that the theory is internally sound, a gap connected to one of the seven Millennium Prize problems: Yang-Mills existence and mass gap.
OpenAI says its models produced a proof of the existence part, first for a simplified version of the theory on September 9, then for the full theory nine days later. The Millennium Prize problem has two components, existence and the “mass gap,” and OpenAI has not claimed to solve the mass gap portion. That means the $1 million prize associated with the full problem has not been claimed. OpenAI also says the work touches on a “completion” of parts of the Standard Model, the framework describing all known particles, though this piece has not been made public. The proof has reportedly been handed to a group of physicists led by Nobel laureate David Gross, who will decide when and how to release it, so independent verification hasn’t happened yet.
Is any of this independently verified? #
No, not fully, and that distinction matters more than the headlines. All three results are described as claims pending outside review. The computer science result is being examined by Ryan Williams alongside the hypothesis’s original authors. The math proof has been posted publicly for any mathematician to check. The physics proof hasn’t even been released publicly yet, pending review by Gross’s group. Reactions from people in these fields have ranged from cautious optimism to outright astonishment, with one researcher calling the complexity theory result possibly the biggest surprise in two decades of theoretical computer science, while others on forums like Hacker News noted these kinds of results can be fragile, with fuzzy boundaries between what’s actually proven and what’s assumed.
Why should non-mathematicians care about this? #
Because the practical effects, if verified, don’t arrive as a single dramatic event. They show up as incremental improvements: cheaper deliveries from better routing algorithms, more efficient computer chips, improved medical imaging, better power grid and hospital scheduling. The exponential time hypothesis result, if it holds, could ripple into any industry that depends on solving large combinatorial problems, which is most of them. The math and physics results are further from immediate consumer impact but reflect progress on foundational tools, wave mathematics and particle physics, that eventually filter into real technology over years, not days.
There’s also a broader signal in the fact that OpenAI is framing this as a preview of how most people will experience advanced AI: not through chatbots writing text, but through accelerated scientific progress that eventually touches everyday products and services.
Frequently Asked Questions #
Did OpenAI’s AI really disprove the exponential time hypothesis?
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
OpenAI claims its model found a proof showing a faster-than-expected method for a class of hard computational problems, challenging a 27-year-old assumption. The claim is being reviewed by outside experts, including the hypothesis’s original authors, and has not been independently confirmed as of the announcement.
Does this mean encryption and internet security are now broken?
No. The claimed result does not prove P equals NP, and Greg Brockman stated that P still probably doesn’t equal NP. Standard encryption methods are not affected by this claim.
What is the local smoothing conjecture?
It’s a 1991 conjecture by mathematician Christopher Sogg about how waves smooth out as they spread, connected to harmonic analysis, a field underlying technologies like MRI imaging and audio compression. OpenAI claims its AI proved it in every dimension at once.
Did OpenAI’s AI win the Yang-Mills Millennium Prize?
No. OpenAI claims a proof of the “existence” portion of the Yang-Mills existence and mass gap problem, but not the “mass gap” portion, which is required to claim the $1 million Millennium Prize. The physics proof also hasn’t been publicly released yet.
Has any of this been confirmed by outside mathematicians or physicists?
Not fully. The math proof has been posted publicly on GitHub for review. The computer science claim is being examined by leading researchers in the field. The physics proof is being reviewed privately by a group led by Nobel laureate David Gross and has not yet been published.