OpenAI GPT-5.2 Derives New Theoretical Physics Result in Landmark AI Discovery

Caroline Bishop Mar 03, 2026 19:59

GPT-5.2 Pro proposed a novel formula for gluon amplitudes that human physicists later verified, marking a first for AI-generated scientific discoveries.

OpenAI's GPT-5.2 Pro has done something no AI system has publicly achieved before: it proposed an original formula in theoretical physics that human researchers subsequently verified as correct. The preprint, published March 3, 2026, demonstrates that certain gluon scattering amplitudes previously assumed to be zero can actually occur under specific conditions.

The discovery challenges textbook assumptions about particle physics. For decades, physicists believed that when one gluon has negative helicity while the remaining particles have positive helicity, the tree-level amplitude must equal zero. GPT-5.2 identified a mathematical exception—the "half-collinear regime"—where this reasoning breaks down and the amplitude becomes nonzero.

How the AI Got There

Human researchers first calculated amplitudes for small particle numbers up to n=6 by hand, producing what the paper describes as "very complicated expressions" whose complexity grows superexponentially. GPT-5.2 Pro simplified these unwieldy formulas dramatically, spotted a pattern across the cases, and proposed a general formula valid for all n.

Here's where it gets interesting: a scaffolded version of GPT-5.2 then spent roughly 12 hours reasoning through the problem independently, arriving at the same formula and generating a formal proof. The equation was verified against the Berends-Giele recursion relation and soft theorem constraints—standard validation methods in the field.

The preprint lists authors from the Institute for Advanced Study, Vanderbilt University, Cambridge, Harvard, and OpenAI. Kevin Weil appears on behalf of OpenAI.

What Physicists Are Saying

Nima Arkani-Hamed, Professor of Physics at the Institute for Advanced Study, called the results "strikingly simple expressions" and noted that finding simple formulas has "always been fiddly" work he long believed could be automated. "I am looking forward to seeing this trend continue towards a general purpose 'simple formula pattern recognition' tool in the near future," he said.

Nathaniel Craig at UC Santa Barbara was more direct about implications: "This is clearly journal-level research advancing the frontiers of theoretical physics." He described the work as "a glimpse into the future of AI-assisted science" and a template for validating LLM-driven insights.

The Bigger Picture

GPT-5.2 launched December 11, 2025, with OpenAI touting massive improvements in reasoning and agentic workflows. The Pro variant specifically takes more compute time to work through complex problems—exactly the capability that apparently enabled this physics breakthrough.

The team has already extended these amplitude calculations from gluons to gravitons using GPT-5.2 assistance, with additional results forthcoming. The preprint is available on arXiv and under review for journal publication.

For AI companies racing to demonstrate real-world value beyond chatbots and code completion, a verified contribution to fundamental physics represents a different category of achievement entirely. Whether this becomes a repeatable methodology or remains a one-off demonstration will likely shape how research institutions approach AI collaboration in the coming years.

• openai
• gpt-5.2
• artificial intelligence
• theoretical physics
• scientific research