A team of theoretical physicists has achieved a breakthrough in particle physics with the help of an advanced AI system, marking what researchers believe is the first major discovery in the field driven by artificial intelligence. The work, led by Harvard’s Andrew Strominger, along with colleagues David Skinner from Cambridge University, Alfredo Guevara from the Institute for Advanced Study, and Alex Lupsasca of OpenAI, resulted in a proof that single-minus gluon tree amplitudes are nonzero—a long-standing puzzle in quantum field theory. The team had been stalled for months trying to verify a conjecture about gluons, particles that carry the strong nuclear force. While they had derived a complex mathematical expression, they couldn’t simplify or confirm it. Earlier attempts to use standard versions of ChatGPT failed, with the AI producing incorrect or incoherent results. But when they tested OpenAI’s internal, more powerful version—dubbed “Super Chat”—the system eventually solved the problem after running for 12 hours. The AI not only proposed a solution but also provided a valid mathematical pathway that the team could verify. After a week of manual checks and refinements, the physicists turned their findings into a preprint paper published on arXiv. The discovery, while technical, represents a turning point in how AI can assist in high-level scientific research. Lupsasca, who joined OpenAI after a promising academic career, had once been skeptical of AI’s capabilities. His perspective shifted dramatically when he used the Pro version of ChatGPT-5.2 to solve a difficult differential equation in astrophysics—something that took him hours—only to have the AI solve it in under 30 minutes using a rare mathematical identity from a 1950s Norwegian journal. That moment, he said, made him “AI-pilled,” leading him to join OpenAI to help improve AI’s reasoning in science. OpenAI responded by launching OpenAI for Science, a new initiative to train AI models in specialized domains like theoretical physics and mathematics. Lupsasca became its first hire, and his collaboration with Strominger helped demonstrate the potential of AI as a true research partner. Strominger described the experience as transformative, noting that working with the AI felt less like interacting with a machine and more like collaborating with a creative mind. He acknowledged that while the team might have eventually solved the problem, they might also have remained stuck indefinitely. The event has sparked broader conversations about the future of science. Rather than replacing human researchers, AI appears to be amplifying their capabilities. As Strominger put it, the key is not fear of obsolescence but the need to retool—adapting to new tools that allow scientists to tackle problems once thought too difficult. This milestone signals a new era in scientific discovery, where AI is no longer just a tool for drafting or data processing but a co-author in the creative process of theoretical physics.