A new artificial intelligence system, named PACMAN, has been developed by researchers at the U.S. Department of Energy's Princeton Plasma Physics Laboratory and Princeton University. This system is designed to monitor and control fusion plasma—hot, ionized gas that could one day power clean energy plants—far more quickly than a human operator. Unlike humans, who may take seconds to react, PACMAN can make decisions in milliseconds, which is crucial for managing the fast-changing conditions inside a fusion reactor. PACMAN was tested on a real fusion system in five experiments, demonstrating its ability to manage complex tasks. The system works by collecting real-time data from a tokamak, a doughnut-shaped device used to contain plasma. This data includes measurements of temperature, density, and magnetic fields. Machine learning models within PACMAN analyze these measurements to predict the plasma's current state or what it might do next. Based on these predictions, the system determines what actions are needed, such as adjusting the power of a heating beam. One of PACMAN’s key features is its ability to resolve conflicting instructions from different controllers and apply strict safety limits before sending commands to the tokamak. During testing, it successfully allowed an AI model to control heating systems, predicted sudden energy bursts, detected and controlled waves in the plasma, and even stopped a potential instability known as a tearing mode before it began. This ability to predict and prevent issues is a major advantage over traditional control systems, which typically only detect problems after they start. PACMAN also showed its versatility by coordinating all six gyrotrons—devices that heat the plasma using high-powered microwave beams—on the DIII-D tokamak. It adjusted both the power levels and the positioning of the gyrotrons’ mirrors in real time to meet complex goals set by researchers. However, the developers stress that PACMAN is not meant to replace human oversight. Human physicists still set the parameters for control, and they review the results after each experiment to refine the system. The modular design of PACMAN could also make it adaptable for other tokamaks, regardless of their size, shape, or instrumentation, even for future fusion machines not yet built.