OpenAI has published a physics paper showing that artificial intelligence can contribute to cutting-edge research in one of the most abstract areas of mathematics, the study of how gravity behaves at the quantum level.
The paper, co-authored by researchers from OpenAI, Harvard University, the University of Cambridge, Vanderbilt University and the Institute for Advanced Study, examines a problem involving gravitons, the theoretical particles that carry the gravitational force.
At issue is a longstanding assumption in particle physics: that a particular class of graviton interactions, known as single-minus amplitudes, always cancel out to zero under standard conditions.
The researchers found an exception.
When the particles involved move in a specific, constrained way relative to one another, the cancellation breaks down and the amplitudes produce well-defined mathematical results, contradicting the general expectation.
The finding connects to a deep symmetry in theoretical physics first identified by the mathematician Roger Penrose, suggesting the result is not a quirk but part of a broader underlying structure.
Related reading
- OpenAI publishes GPT-5.3 Instant system card
- OpenAI strikes classified AI deal with US military and calls for industry-wide terms
- OpenAI and Amazon announce strategic partnership
OpenAI said its GPT-5.2 Pro model assisted parts of the derivation, with results verified analytically and checked against known physical limits.
The paper has been posted as a preprint, meaning it has not yet undergone formal peer review, and OpenAI said it welcomed feedback from the scientific community.
The recap
Preprint shows single-minus graviton tree amplitudes are nonzero.
Results arise in a half-collinear kinematic regime of momentum space.
Authors say further extensions are currently under investigation.