OpenAI has published a new research preprint detailing what it describes as AI’s first original contribution to theoretical physics. According to the company, a specialized research version of GPT-5.2 independently derived a mathematical formula in particle physics and formally proved its correctness.

Challenging a “Solved” Result

The paper focuses on a problem long assumed to be settled in the literature. GPT-5.2 reportedly determined that the accepted solution was incorrect and proposed a revised formulation, accompanied by a full mathematical proof. OpenAI says the system autonomously generated the proof within roughly 12 hours of compute time.

Academic Verification

The work was reviewed and verified by physicists from institutions including Harvard, Cambridge, and Princeton. OpenAI researcher Kevin Weil is listed as a co-author, alongside Harvard physicist Andrew Strominger, who said the AI “chose a path no human would have tried,” highlighting the unconventional reasoning route taken by the model.

From Tool to Research Contributor?

While AI systems have assisted with literature reviews and equation manipulation before, OpenAI positions this result as qualitatively different: an autonomous derivation that challenges an established answer and withstands expert scrutiny. Skeptics are likely to debate whether the contribution qualifies as “original” in a human sense, but the development underscores how frontier models are increasingly being directed at open scientific problems rather than productivity tasks.

Why it matters: If independently verified over time, this would represent a shift from AI as a research assistant to AI as a research collaborator — capable not only of accelerating known workflows, but of interrogating and correcting long-standing theoretical assumptions in high-level science.