On the margins of the 2025 AI for Good Summit, the Simon Institute for Longterm Governance (SI) co-hosted a Geneva Security Debate on the idea of a “CERN for AI” with the Geneva Centre for Security Policy (GCSP). CERN, the European Organization for Nuclear Research, was founded in Geneva in 1954 as a joint European project to fund fundamental, civilian, and open research in particle physics. The analogy of a “CERN for AI”, first coined by Prof. Gary Marcus at the 2017 AI for Good Summit, evokes a vision of cooperative, public-interest-driven AI development.
As we highlighted in a blog post prior to the event, there are still a range of different ideas around a CERN for AI, and there is a need for more strategic clarity on the vision. The event aimed to contribute to this by bringing together Prof. Gary Marcus (cognitive scientist, entrepreneur, & author), Prof. Mary-Anne Hartley (EPFL / ICAIN), Balint Pataki (Centre for Future Generations), Chiara Gerosa (Talos Network), Kevin Kohler (SI), and Dr. Christina Schori Liang (GCSP) who moderated the event. Although many questions remain unanswered, the following points represent some common ground:
1. A CERN for AI needs an ambitious, inspiring mission
Panelists emphasized that talent isn’t solely drawn by salaries. A CERN for AI needs a clear mission for the public good. Examples like India’s Aadhaar system or the UK’s AI Security Institute show that visionary public projects can rally top-tier technical talent even in competition with the giant salaries from tech giants, if they prioritize mission and impact over bureaucratic processes. If we create artificial general intelligence, solely using it to addict teens to chatbots rather than to solve humanity’s grand challenges would seem like an absurd misuse of potential.
2. A CERN for AI needs strong talent clustering
Top researchers want to work with other top researchers. Unlike the talent concentration in Silicon Valley, European AI talent is more fragmented across hubs like Zurich, Paris, Berlin, and Barcelona. Speakers echoed calls from the Centre for Future Generations and CLAIRE/CAIRNE to consolidate research efforts across Europe in order to foster serendipitous innovation. As highlighted by the Centre for Future Generations, this is compatible with multiple, geographically separate AI Gigafactories.
3. The right time to set up a CERN for AI style institution is now
Technology is moving fast and we are in a challenging geopolitical environment. While some participants highlighted that Silicon Valley’s stated AGI timelines require a cautious and critical perspective, others underscored that AI risks are increasingly relevant either way, including misinformation and truth decay. Overall, there was a shared sense of urgency about the narrowing window of opportunity for a coalition of actors to advance a more collaborative, globally oriented vision of AI for the public good.
It will take several years for Europe’s new AI Gigafactories to become operational. However, a CERN for AI could start its mission immediately by allocating GPU hours from public and commercial clouds to researchers. Similarly, while it would be impossible to create a “de novo” AI talent hub within a brief period of time, an existing AI talent ecosystem can be chosen as a starting point for a CERN for AI hub.
A recording of the event by GCSP will soon be available here.
Photo credit: GCSP