AI Scientist Spots What Physicists Missed in Gluon Scattering

A groundbreaking study reveals AI's role in discovering a new formula governing gluon interactions, challenging long-held assumptions in particle physics.

Published on Feb. 14, 2026

Got story updates? Submit your updates here. ›

A new study published on arXiv demonstrates how artificial intelligence, specifically the GPT-5.2 Pro system, was able to identify a previously unknown formula governing the interactions of gluons - fundamental particles that mediate the strong force. Conventional understanding suggested certain gluon interactions should 'vanish' at the most basic level of calculation, but the AI was able to simplify complex expressions, identify a pattern, and propose a general formula that was then independently verified. This discovery not only advances theoretical physics but also points to a potential paradigm shift in how scientific research is conducted, with AI playing a key role in generating hypotheses and proofs.

Why it matters

This study highlights the growing potential for AI systems to make significant contributions to fields like particle physics, where identifying hidden patterns and simplifying complex mathematical expressions has traditionally been a painstaking, human-driven process. The ability of AI to autonomously propose and validate new theoretical formulas challenges long-held assumptions in the field and opens up new avenues for exploration, particularly in areas like graviton interactions and supersymmetric extensions of the Standard Model.

The details

Gluons possess a property called 'helicity' that describes the orientation of their spin. Conventional understanding suggested that certain interactions between gluons, specifically 'single-minus amplitudes,' should vanish at the most basic level of calculation. However, researchers found that under specific conditions, these interactions do exist and can be expressed with a remarkably simple mathematical formula. The key to this discovery was the GPT-5.2 Pro AI system, developed by OpenAI. Scientists initially calculated amplitudes for small values, resulting in complex expressions. The AI then simplified these expressions, identified a pattern, and proposed a general formula applicable to all values. Remarkably, the AI then spent approximately 12 hours independently verifying its own conjecture, producing a formal proof before human researchers confirmed the result analytically.

  • The research was detailed in a preprint published on arXiv on February 13, 2026.

The players

GPT-5.2 Pro

An AI system developed by OpenAI that was instrumental in identifying a new formula governing gluon interactions.

Nima Arkani-Hamed

Professor of Physics at the Institute for Advanced Study, who noted that finding simple formulas in this area of physics has always been a fiddly process, and one that seemed ripe for automation.

Nathaniel Craig

Professor of Physics at the University of California, Santa Barbara, who highlighted the potential for AI to generate 'fundamentally new knowledge' when coupled with human expertise.

Got photos? Submit your photos here. ›

What they’re saying

“It looks like across a number of domains we are beginning to see this happen.”

— Nima Arkani-Hamed, Professor of Physics at the Institute for Advanced Study (OpenAI blog post)

“a glimpse into the future of AI-assisted science.”

— Nathaniel Craig, Professor of Physics at the University of California, Santa Barbara (Preprint)

What’s next

The researchers suggest the same principles could apply to gravitons - hypothetical particles mediating gravity - and supersymmetric extensions of the Standard Model. The study also opens doors to exploring more complex scenarios, including loop corrections which account for quantum fluctuations.

The takeaway

This study demonstrates the potential for AI systems to revolutionize scientific discovery, not just in data analysis but in generating hypotheses and proofs. The ability of AI to identify hidden patterns and simplify complex mathematical expressions could lead to breakthroughs across various fields of physics, challenging long-held assumptions and paving the way for a new era of AI-assisted scientific research.