Fermilab-Led MOAT Collaboration Unveils AI Tool That Speeds Accelerator Tasks 100-Fold

Particle accelerators are machines that propel charged particles to high energies, enabling research in medicine, energy, and environmental cleanup. They contain tens of thousands of interconnected components, making their design and operation highly complex.

Jonathan Jarvis, MOAT collaborator and director of Fermilab’s Accelerator Research Division, said the goal is to integrate AI so fully into accelerator design, construction, and operations that the pace of discovery and resulting innovations are fundamentally transformed. The Osprey AI tool demonstrated by MOAT can accelerate specific accelerator tasks up to 100 times faster. Researchers from Berkeley Lab, Argonne National Laboratory, Jefferson Lab, Oak Ridge National Laboratory, SLAC National Accelerator Laboratory, and Brookhaven National Laboratory are collaborating with Fermilab on the MOAT AI integration project.

By using AI agents that reason and act autonomously, Osprey reduces the time needed for tasks such as beam tuning and error diagnosis. MOAT is also developing digital twins—virtual replicas of accelerator complexes—to test changes before applying them to hardware, which could save time and cost.