Strengthening European AI Safety Unigroups

Europe has a strong talent pool, but compared to the US or UK only very few AI safety university groups. This project maps where European groups are, what they are doing, and what would help them most, so that programmes like Pathfinder and national groups can support them better. The ideal outcome is a playbook for starting and running European AI safety university groups, useful both for existing organisers and for people thinking about starting something new.

Possible questions to investigate:

• How do differences between university systems shape the way a group gets started, given that clubs are far less common in many European countries?
• Where do groups exist today, how large are they, how active are they, and what are they actually doing?
• What has worked well already, and what would organisers have wanted to know when they started?
• What local vision should a group work with, and how much should it be a general AI safety pipeline versus focused on local levers?
• At which universities would a new group be most valuable, and what would a founder there need?
• How can we improve connections between US, UK and European uni groups?

Pathfinder and most existing organiser resources are centred around US and UK universities and often do not transfer well to EU groups. This project works out what European organisers actually need, drawing on what groups here have already figured out, so that Pathfinder and national groups can support existing groups and seed new ones more effectively. Through this more talent in Europe starts working in AI safety and uni groups become a real talent pipeline across the continent.

There are two main streams this project can go.

Landscape and seeding: Map the current state of European AI safety uni-groups: which exist, how large and active they are, what programming they run, how they are funded, how they sit within their university and how well they are connected to other groups in their country and the EU. For this, you interview current and potentially past organisers of groups and write up your findings as a community resource. From there, you work out where new groups would be particularly valuable and what a founder there would need.

Local visions and curricula: Most existing curricula teach AI safety in general, and there is a lot of value in going one step further, so that members also understand which role their own country could play and what they could realistically do from there. That means interviewing organisers and people who know a given national context well, and turning what they say into short country curriculum additions that an organiser can pick up and use.

Both streams can come together in a playbook, which should be modular and written for someone deciding whether to start or restart a group. Alongside the playbook, we would create a memo for Pathfinder and national groups on what organisers told us they need. As a stretch, we could try to approach potential founders at some of the universities we would like a group to exist at.

• Europe-based, or recently studied at a European university
• Comfortable with cold outreach and interviewing people you do not know, since this is most of the work
• Enough AI safety context to hold a good conversation with an organiser or researcher (BlueDot or AI Safety Atlas)
• Speaks at least one European language besides English
• Having run a group yourself is a plus, but not required

Europe-based, or recently studied at a European university.