A new Europe-wide research network aims to bring order — and trust — to the rapidly growing use of artificial intelligence in fundamental physics. The AIFORIA network (Artificial Intelligence for Fundamental Physics: Open Roadmapping and Inclusive Actions) officially launched on 1 November 2026 and will run until 2030, according to the University of Amsterdam (UvA).

Christoph Weniger, a physicist at the UvA Institute of Physics and GRAPPA, will chair the initiative. Verena Kain of CERN serves as Vice-Chair. The network brings together researchers from 28 countries and is organised as a COST Action — a European framework that funds coordination, meetings and training across borders — under the umbrella of the European Coalition for AI in Fundamental Physics (EuCAIF), a bottom-up community that Weniger co-founded.

AI is already embedded in day-to-day physics research, the UvA reported. Algorithms help identify rare collision events at the Large Hadron Collider, pick out gravitational-wave signals from noisy detector data, and compare cosmological simulations with telescope observations. But each subfield — particle physics, astroparticle physics, nuclear physics, gravitational physics and cosmology — has largely developed its own methods and tools in isolation. A central unresolved question, the university noted, is when AI results can be trusted. Physics, Weniger and his colleagues argue, needs methods that can state their own uncertainty and that respect known physical laws.

The challenge will only grow. Upcoming facilities such as the Einstein Telescope, the LISA space mission and the Square Kilometre Array are expected to produce vastly more data than current instruments, making automated analysis essential.

AIFORIA is organised into three working groups. The first focuses on methods: which AI techniques work best for scientific discovery, to be tested on common benchmark problems so they can be compared fairly. The second examines the environment these methods run in — shared software, data and computing resources — as well as the rise of agentic AI systems that can carry out parts of the research workflow autonomously. The third group, working with EuCAIF, will develop a joint European vision for AI in fundamental physics over the coming years and train the next generation of researchers through PhD schools.

The working groups will hold a joint kick-off meeting in early 2027, and the first PhD school is planned for summer 2027, the UvA said. Young researchers and scientists from countries that are less represented in European research will receive particular support.

“AI systems are starting to carry out parts of the research process themselves, and much of that development happens outside Europe,” Weniger said in the university’s announcement. “Nobody can predict where this will be in five years, but as a community we can agree, from the bottom up, on what we want to get right.”

The network supports hackathons, short research exchanges and other collaborative activities. Researchers and innovators from academia and industry can join at any time by creating a free e-COST account and applying to one or more working groups via the COST website. More information is expected at aiforia.eucaif.org in the coming weeks.

For university students considering careers in physics, data science or AI, the network signals that European research is making a coordinated push to shape how AI is used in science — and that training opportunities, including PhD schools and exchanges, will be available across the continent over the next four years.