What happened?
In February 2026, five authors posted a preprint arguing that a type of gluon interaction usually assumed to be zero can be nonzero in a special situation arXiv preprint ↗. Gluons are the particles that hold quarks together inside protons. OpenAI reports that GPT-5.2 Pro proposed the general formula and a newer internal OpenAI model proved it OpenAI blog ↗.
Physicists describe how particles collide using scattering amplitudes, numbers that give the chance of each outcome. Each gluon has a helicity, a kind of spin direction that is either plus or minus. A textbook argument says the simplest amplitude with one minus gluon and all others plus must be zero arXiv preprint ↗. The authors found that this argument does not cover a narrow set of momentum arrangements they call half collinear, where the amplitude can be nonzero arXiv preprint ↗.
The setting is unusual. The half collinear arrangement is possible in a mathematical version of spacetime called Klein space, or with complex number momenta, but not in the ordinary spacetime of everyday physics arXiv preprint ↗ Quantum Insider ↗. The result is also at tree level, the simplest level of calculation; harder corrections were not covered Quantum Insider ↗. The authors are from the Institute for Advanced Study, Vanderbilt, Cambridge, Harvard and OpenAI OpenAI blog ↗.
Who did which step?
Humans calculated
The physicists worked out cases with up to six gluons by hand, which gave very long expressions OpenAI blog ↗.
GPT-5.2 Pro guessed
It simplified those expressions, spotted a pattern and proposed a formula for any number of gluons OpenAI blog ↗.
An internal model proved
A scaffolded internal version of GPT-5.2 spent about 12 hours and produced a proof, which the authors then checked OpenAI blog ↗.
An AI spotted a simple pattern inside messy hand calculations, and that pattern passed standard physics checks. Finding simple formulas behind complicated ones is a common route to progress in theoretical physics.
Leapscope interpretation of the reported result.How did AI help?
The human physicists chose the question, did the first calculations and checked the result OpenAI blog ↗. The paper itself says the formula was "first conjectured by GPT-5.2 Pro and then proved by a new internal OpenAI model" and was checked by hand against a standard recursion method, called Berends-Giele recursion arXiv preprint ↗. The formula also passed checks against symmetry rules and Weinberg's soft theorem, a known rule for low energy particles Quantum Insider ↗.
Both figures reported by OpenAI OpenAI blog ↗.
The description of the AI's role comes from OpenAI and the authors, two of whom work at OpenAI OpenAI blog ↗. The internal model that wrote the proof is not publicly available and its proof is not described in detail on OpenAI's page OpenAI blog ↗. Two outside physicists praised the work, but their comments were published by OpenAI and are not formal peer review Quantum Insider ↗.
Which fields could this affect?
The immediate value is to theoretical physics research; wider effects are speculative. These connections are our assessment.
Theoretical particle physics
Researchers can study and test the formula in the public preprint arXiv preprint ↗. A follow up preprint reports a similar result for gravitons, particles of gravity in theory Graviton follow up ↗.
Explore scienceAI tools for physicists
The case suggests AI can help find simple patterns in long calculations Quantum Insider ↗. Whether that works on other problems is untested.
Explore softwareUnderstanding real collisions
The result holds in a special mathematical setting, not ordinary spacetime Quantum Insider ↗. It has no direct link to experiments or energy technology.
Explore scienceWhat has been checked?
The evidence is an arXiv preprint and the developer's announcement, with news coverage that relies partly on it. Leapscope reviewed these sources; we did not repeat the calculations.
Shown so far
- The authors derived a closed form formula for the case of one minus gluon decaying into many plus gluons arXiv preprint ↗.
- The formula passed checks against Berends-Giele recursion and Weinberg's soft theorem arXiv preprint ↗ Quantum Insider ↗.
- GPT-5.2 Pro proposed the general formula, according to the paper and OpenAI OpenAI blog ↗ arXiv preprint ↗.
Still unknown
- Whether peer reviewers and outside physicists accept the result.
- How the internal model's proof was produced, since that model is not public.
- What role these amplitudes play in the wider theory Quantum Insider ↗.
Evidence status: Preprint; checked by authors. Stage: Claim. A preprint. Peer review and independent checks are pending.
From preprint to accepted physics
This is our suggested way to follow the story, not a promised timetable.
Can I use it today?
The preprint is free on arXiv for anyone to read arXiv preprint ↗. GPT-5.2 Pro is an OpenAI product, but the internal model that wrote the proof is not available OpenAI blog ↗. There is no practical device or technology from this result.
A few things you might be wondering
Did AI discover new physics on its own?
No. Human physicists set up the problem and did the first calculations; the AI guessed and proved a general formula, and the humans checked it OpenAI blog ↗.
Does this change what happens in particle colliders?
Not directly. The result applies to a special momentum arrangement that is not possible in ordinary spacetime Quantum Insider ↗.
Has it been peer reviewed?
Not yet. It is an arXiv preprint, updated in March 2026, with no journal publication listed arXiv preprint ↗.
Go straight to the sources
Checked Oct 8, 2026. The first source is the original announcement or research. Later sources add independent context; background pages do not validate the result on their own.
01OpenAI's account of how GPT-5.2 and an internal model contributed, with outside comments.
The paper by Guevara, Lupsasca, Skinner, Strominger and Weil.
News report explaining the result and its limits.
Follow up applying the same idea to gravitons with GPT-5.2 Pro's help.