A new artificial intelligence system called HEAT-ML has dramatically accelerated the process of identifying safe zones, or "magnetic shadows," inside fusion reactors. Developed through a public-private partnership between Commonwealth Fusion Systems (CFS), the U.S. Department of Energy’s Princeton Plasma Physics Laboratory (PPPL), and Oak Ridge National Laboratory, the AI tool is designed to predict where intense plasma heat will strike a tokamak’s internal components—and where it will be shielded. These magnetic shadows are crucial for protecting the reactor’s walls and components from melting due to plasma temperatures that exceed those at the core of the sun. Without accurate predictions, fusion experiments could face costly shutdowns or damage to critical hardware. HEAT-ML was specifically developed to support the design and operation of SPARC, a compact tokamak under construction by CFS with the goal of achieving net energy gain by 2027. The system focuses on a high-stress region of SPARC’s exhaust system—15 tiles near the bottom of the machine—where the most extreme heat fluxes occur. Traditionally, researchers used an open-source code called HEAT (Heat flux Engineering Analysis Toolkit) to generate 3D shadow masks that map these protected zones. However, simulating these masks was computationally intensive, often taking up to 30 minutes per run and longer for complex geometries. This delay hindered rapid design iterations and real-time decision-making during experiments. HEAT-ML overcomes this limitation by using a deep neural network trained on about 1,000 simulations generated by the original HEAT code. The AI learns to predict shadow masks by analyzing patterns in magnetic field lines and their interactions with the tokamak’s 3D structure. Once trained, HEAT-ML can produce results in just a few milliseconds—speeding up the process by thousands of times. While currently tailored to SPARC’s specific exhaust design, the research team aims to expand HEAT-ML’s capabilities to work across different tokamak geometries and plasma-facing components. The ultimate goal is to create a flexible AI-driven tool that can support the design and real-time control of future fusion reactors. Michael Churchill, co-author of the study and head of digital engineering at PPPL, said the work demonstrates how AI surrogates can enhance scientific computing by turning slow simulations into fast, actionable insights. He highlighted the potential for AI to improve scenario planning and operational control in fusion experiments. Doménica Corona Rivera, the paper’s first author and an associate research physicist at PPPL, emphasized the importance of preventing plasma contact with reactor walls. “The worst thing that can happen is that you would have to stop operations,” she said. The project was supported by the U.S. Department of Energy under contracts DE-AC02-09CH11466 and DE-AC05-00OR22725, with additional backing from Commonwealth Fusion Systems. The success of HEAT-ML marks a significant step toward making fusion energy a practical and reliable power source.