The forgotten science15 minBefore 'AI' existed, a group of mathematicians, engineers, doctors and anthropologists invented cybernetics: the science of feedback, control and information. It gave us the computer's blueprint, the first artificial neurons and the math behind chatbots. Its founder also predicted the problem AI labs now call alignment.Read now →

An AI Just Put the Force Holding Every Atom Together on Solid Math. Its Own Referees Admit They Don't Fully Understand the Proof
OpenAI says its model proved that Yang-Mills theory, the math behind the strong nuclear force, actually exists. A committee led by Nobel laureate David Gross has now validated the computer-checked version. Here's what was proved, what wasn't (the $1 million 'mass gap' is still open), and why the way it was released has mathematicians angry.
Almost all of your body's mass doesn't come from "stuff."
It comes from energy: the frantic motion of particles called gluons, buzzing inside every proton and neutron in every atom in you, right now. The theory that describes that force is one of the most successful in science. It's been confirmed by mountains of experiments.
And for half a century, nobody could prove it was mathematically coherent.
This month, that changed. OpenAI says one of its AI models produced a proof that the theory, known as Yang-Mills theory, really exists in the rigorous sense mathematicians demand. On November 10, 2026, a committee led by Nobel laureate David Gross released its validation of a computer-checked version of the proof.
It's a landmark. It's also only half of one of the most famous unsolved problems in mathematics, and the way it was released has started a fight about how AI-generated science should be shared.
Here's the whole story, in plain language.
The force that holds atoms together
Look inside a proton and you'll find three quarks. Add up their masses and you get only about 1% of the proton's mass.
So where's the other 99%?
The strong nuclear force is carried by particles called gluons, much as light is carried by photons. Quarks constantly exchange gluons. Gluons themselves are massless, but the energy of that frantic activity contributes mass through Einstein's E = mc². That energy accounts for nearly all of the proton's mass, and the same goes for neutrons.
Since almost all your mass is in protons and neutrons, almost all of you is, in a real sense, the energy of the strong force.
The theory describing it is a kind of Yang-Mills theory, developed by physicists Chen Ning Yang and Robert Mills in 1954. Yang-Mills theories underpin both the strong force and the weak nuclear force, and they're a central pillar of the Standard Model of particle physics.
The problem: infinities
Here's the awkward secret of quantum field theory.
In the middle of calculations, physicists regularly hit results that would make a traditional mathematician's ears smoke: infinite energies, infinite probabilities. In a quantum field, even empty space fizzes with fluctuations, and at very small scales those fluctuations can involve arbitrarily high energies.
Physicists have found clever workarounds to tame these infinities. Two classic tricks:
- Pretend space is a grid. Instead of infinitely many points between two locations, use a fine lattice.
- Pretend space is a finite box. Instead of space going on forever, put everything inside a container.
With those assumptions, the infinities behave, you get finite answers, and they match experiment beautifully.
But space, as far as we know, isn't a grid inside a box. So the deep question is: can you gradually remove the grid and the box and still have a mathematically sensible theory? Or do the workarounds paper over something fundamentally broken?
That's a question about whether the theory exists at all, in the rigorous mathematical sense.
A million-dollar problem, in two halves
In 2000, the Clay Mathematics Institute named seven Millennium Prize Problems, each with a $1 million reward. One is "Yang-Mills Existence and Mass Gap." It has two parts:
- Existence: prove that quantum Yang-Mills theory can be rigorously constructed in continuous, infinite, four-dimensional spacetime, satisfying the standard mathematical axioms for a quantum field theory.
- Mass gap: prove that the theory has a mass gap, meaning every particle it produces has mass, with a minimum energy above empty space.
As Scientific American put it, the second part is "more quantitative (and far more difficult)."
OpenAI's result addresses the first part only.
What "existence" really means
"Proving a theory exists" sounds absurd. Physicists use it every day. Of course it exists.
Here's an analogy. Imagine a recipe that has always produced a perfect cake, but one step says "add an infinite amount of flour, then remove an infinite amount of flour." Bakers learned that if they just skip both steps, the cake comes out fine. Nobody ever checked whether the recipe, as written, describes anything real.
Physics' workarounds are a bit like that. They give right answers. The existence question asks whether there's a genuine, well-defined mathematical object underneath those answers, one that satisfies a precise list of rules (the axioms a quantum field theory is supposed to obey), with no grid and no box. Until now, for four-dimensional Yang-Mills, nobody could construct one.
What OpenAI announced
The timeline, according to reporting in Science and Scientific American:
- October 28, 2026: OpenAI released a batch of AI-generated results, including a proof that pure Yang-Mills theory exists in four dimensions, plus a refutation of a decades-old conjecture in computer science.
- Instead of publishing the proof, which runs to 1,147 pages, OpenAI appointed a committee of physicists and mathematicians, led by David Gross, to "shepherd" the result.
- November 9: the committee previewed its findings, saying OpenAI's result "proves the existence of pure Yang-Mills theory according to the problem statements which are standard in physics and mathematics."
- November 10: the committee released a document explaining how the proof works, alongside its validation of a formalized version in Lean, a programming language that mechanically checks every logical step against a set of axioms.
An independent assessment of the construction describes it as establishing a four-dimensional pure Yang-Mills quantum field theory that satisfies the standard Wightman axioms, built as a continuum limit of lattice gauge theory: literally, taking the grid away.
Harvard physicist Michael R. Douglas, a committee member, summarized the stakes: despite its centrality to physics, four-dimensional Yang-Mills theory "has never been properly understood mathematically, even after more than fifty years of work... No rigorous definition of it existed... Not any more."
"We do not claim to fully understand the proof yet"
Here's the line that stops people cold. In their own assessment, the committee wrote that they used OpenAI's Astra model to help digest the Lean code, and: "We do not claim to fully understand the proof yet."
That's the strange new reality of AI mathematics. The proof is formally checked. The referees vouch for the checking. But no human yet has a deep intuitive grasp of why it works.
Why Lean matters so much here
A proof assistant like Lean works like an extremely strict compiler for mathematics. Every definition must be stated precisely, and every step must follow from earlier steps or accepted axioms. If anything is hand-waved, the check fails.
That's why the committee's validation carries weight even though few humans have read all 1,147 pages: the logic has been checked mechanically. The remaining human job is to confirm that the formal statement in Lean really matches the famous problem as mathematicians state it, and that no hidden assumption slipped in. That's exactly what the wider community can now start doing with the released code.
How the AI did it: finishing a 40-year-old job
This wasn't a bolt from the blue. Both OpenAI and a rival effort built on the work of mathematician Tadeusz Bałaban, who came close to solving the existence problem in the 1980s.
Bałaban's strategy was the grid approach: chop spacetime into a lattice, make the calculations manageable, then show the theory stays consistent as the grid gets ever finer, approaching real, continuous spacetime. He rigorously handled lower-dimensional cases, but the full four-dimensional case remained unfinished.
The obstacle was sheer labor. As University of Helsinki mathematician Antti Kupiainen told Science, Bałaban himself admitted there were complicated calculations still to do, and "it's just too painful for a human to carry that out."
That's exactly the kind of problem AI is good at: a plausible strategy blocked by an overwhelming amount of careful work. Northwestern mathematician Ethan Sussman put it this way: if the obstruction has been "the amount of work involved, that's a pretty decent thing to try to send the AI model towards."
The rival effort
OpenAI wasn't alone. Sussman, working with Anthropic's Levent Alpöge, independently disclosed AI-assisted progress on the same problem, using Anthropic's Claude. Sussman told Science that Claude didn't simply finish Bałaban's arithmetic; it found new ways to organize the proof and address gaps Bałaban had left, including how particles and fields influence one another.
Unlike OpenAI, Sussman and Alpöge published their working process openly on GitHub, and Sussman is translating the AI-generated math into a manuscript humans can follow.
The mass gap: still wide open
Now the part that hasn't been solved.
Remember that gluons, unlike photons, feel the force they carry. Photons are electrically neutral, so they don't attract each other. Gluons do attract each other. Push the math through, and you find a gluon can never fly off on its own. Instead, gluons clump into bound states physicists call glueballs, and because of the energy packed inside, every glueball has mass.
So even though the theory is written in terms of massless gluons, nothing massless ever comes out. Every particle the strong force can produce has some minimum mass. That minimum is the mass gap.
Physicists are almost certain it exists. Powerful computer simulations show it plainly, and it's believed to explain why the strong force is short-ranged even though gluons are massless. As Cambridge physicist David Tong put it, the theory seemingly "somehow conjure[s] a mass out of nothing."
But a simulation isn't a proof. Nobody has proven the mass gap, including AI. The committee itself says it remains open. Stanford's Sourav Chatterjee told Science that mathematicians have no clear path to it yet, though OpenAI's October batch did include work establishing a mass gap in a simpler two-dimensional toy model.
Will OpenAI get the $1 million? Unlikely on this alone: the existence result is only half the problem. (And as more than one observer has noted, the prize money is pocket change to an AI lab.)
The fight over how it was released
The math is impressive. The release strategy is controversial.
- The proof itself wasn't released. The committee validated the Lean formalization, which is strong evidence of logical consistency, but as Science noted, that doesn't by itself mean the full unreleased proof is correct as written.
- Transparency advocates object. The Advisory Group on Mathematics and Artificial Intelligence, hosted at the Institute for Advanced Study, called for the proof's immediate release.
- "Needless gatekeeping." NYU mathematician Scott Armstrong told Scientific American he was glad the Lean proof was finally out and hoped labs would share results widely: "This review can be done more efficiently and quickly in the open."
- The choice of referee raised eyebrows. EPFL mathematician Martin Hairer questioned picking Gross, who had previously dismissed efforts to put quantum field theories on rigorous footing as "useless" and "ugly."
Context matters here. OpenAI's shepherding approach was partly a response to its October "slop drop," when it released more than 700 AI-generated proofs in a single day, overwhelming reviewers. The committee model was meant to avoid a repeat. Critics argue it swapped one problem for another.
Why this matters
For physics: probably not much, day to day
Physicists have used the workarounds successfully for decades, and the Standard Model's predictions don't change. A particle physicist running simulations at a supercomputer center tomorrow will do exactly what she did yesterday. But Tong suggests a rigorous foundation might still reveal something new about deep mysteries like where proton mass comes from or why quarks are never found alone. "Maybe there's something we're missing."
For mathematics: a landmark
It brings mathematicians a big step closer to another Millennium Problem and comes after OpenAI's claims on the Navier-Stokes problem in September and a major advance related to the Riemann hypothesis in October. (We covered the Riemann-adjacent work in our piece on AI and the 7/8 result.)
For how science works
This may be the most lasting change. Sussman hopes AI will re-center what mathematicians consider important. Elegant problems have always attracted talent; grinding, enormous projects like Bałaban's languished. If AI shoulders the labor, neglected but physically important questions become attractive again.
But there's a new problem: reading the output. AI proofs can be hundreds of pages long, with strange terminology for familiar ideas. "The only way to read one of these AI-written manuscripts is with AI," Sussman says.
Practical takeaways
- Separate "proved" from "solved." The existence half is validated; the Millennium Problem isn't solved.
- "Formally verified" is strong but not the whole story. Lean checks logic against axioms; humans still need to confirm the formal statement matches the intended problem, and to understand the ideas.
- Watch for open versions. Sussman's human-readable manuscript and community scrutiny of the released Lean code will determine how quickly this result is absorbed.
- Expect more. The pattern of AI finishing heavy, half-built human programs is likely to repeat across math and physics.
Counterpoints and risks
- Understanding lags verification. A proof no one fully understands is a strange foundation to build on, at least for now.
- Selective release undermines trust. When a company controls access and picks the referees, outsiders can't easily check the claims.
- Hype risk. "AI solves Millennium Problem" headlines overstate what happened; the harder half remains.
- Attribution questions. Bałaban's decades of groundwork, and the parallel Anthropic-linked effort, complicate any single-winner story.
FAQ
Did AI solve a Millennium Prize Problem? Not fully. OpenAI's result addresses the existence half of the Yang-Mills problem. The mass gap half remains unsolved.
Who checked the proof? A committee led by Nobel laureate David Gross validated a formalized version in the Lean proof assistant and released its assessment on November 10, 2026.
What is the mass gap? The claim that every particle produced by Yang-Mills theory has mass, with a minimum energy above the vacuum, which explains why the strong force is short-ranged.
Why are some mathematicians upset? OpenAI didn't release the full 1,147-page proof publicly and instead used an appointed committee, which critics call unnecessary gatekeeping.
The bottom line
For fifty years, the theory describing the force inside every atom worked brilliantly in practice and remained shaky in principle. An AI model has now, by the best available checks, put half of it on solid mathematical ground, by finishing a job a human mathematician started in the 1980s and found too painful to complete.
The hardest part, the mass gap, is still waiting. And the most unsettling detail isn't in the physics at all. It's the referees' honest admission that they verified a result they don't yet fully understand.
That's the new shape of discovery: machines doing the climb, and humans racing to understand the view.
Sources
- Scientific American: Scientists confirm OpenAI's blockbuster quantum physics result
- Science: AI rivals tackle one of physics' deepest mathematical challenges
- Validation of the OpenAI Yang-Mills construction (committee assessment, PDF)
- Yang-Mills group: OpenAI-YM-20261028 repository
- India Today: Nobel laureate validates OpenAI's AI-generated Yang-Mills proof, but $1 million problem remains open
- Clay Mathematics Institute: Yang-Mills and the Mass Gap





