AI agents reportedly found new answers to a fusion-reactor magnetic-field problem

A post claims two papers in September 2026 overturned physicist Harold Grad’s 1967 conjecture that magnetic fields inside twisty fusion reactors could not be solved exactly. It says one paper used OpenAI’s GPT-6 Astra, and Google DeepMind then gave that paper to Argon agents, which found a new family of answers. The post says promising candidates appeared within hours and scientists checked every step by hand.