A US Department of Energy award will expand a Berkeley Lab-led AI project to 16 collaborating institutions, with work planned to bring its accelerator assistant to more facilities. The team also plans to connect the tool to digital twins and use it to help design future accelerators, a practical test of AI in scientific infrastructure rather than another chatbot demo.
Watch Desk analysis
What happened
Berkeley Lab says the Multi-Office Accelerator Team Core Project, or MOAT-Core, is part of Phase II of the DOE’s Genesis Mission. The multi-year project builds on Osprey, an AI assistant that the lab says has been prototyped at the Advanced Light Source. Operators can use natural language to set up experiments, adjust particle-beam settings, troubleshoot equipment and record and synthesise data.
The next phase is intended to deploy Osprey at additional accelerators across the DOE complex and connect its capabilities to digital twins and accelerator design. The lab says the project’s partners include eight national laboratories, universities and industry organisations. Read Berkeley Lab’s announcement.
Why it matters
Accelerators are specialised machines, and sharing software between facilities is not straightforward. A tool that can adapt to different instruments could help researchers reuse work across sites, while AI-assisted design could help narrow a vast range of possible components and configurations. The lab says that process currently takes years of computations and testing; the announcement sets out the ambition, not evidence that the new design approach has already delivered faster or better machines.
Our read
This is an unusually grounded AI project: the aim is to help run and design real scientific equipment, with further deployments still to come. The interesting test is whether Osprey can travel between facilities without becoming a bespoke engineering project at every stop. If it can, the benefits may reach beyond accelerator physics to the science and industrial work those machines support.
What to watch
- Which additional accelerator facilities deploy Osprey, and when.
- Whether its digital-twin and design work produces useful results beyond the prototype.
- How the collaboration handles the different instruments and constraints at each facility.
Discussion spark: If Osprey works across more accelerator facilities, should its developers prioritise a shared platform that many labs can adapt, or bespoke tools tuned to each machine?
Sources and evidence
- Super Intelligence for Particle Accelerators Gets Another Boost – Berkeley Lab News Center (.gov) (8 October 2026, 17:26 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.