Foresight Institute AI Nodes

Context

Funding alone isn’t enough for novel AI safety projects to scale. We have been funding frontier technology and science for 40 years and have seen first hand the need for sustained org support, shared infrastructure, and mission-aligned community to build in. The shortening timelines and consolidation of the funding landscape risks crowding out independent AI safety approaches. To counterweight some of these forces, Foresight launched two AI Safety Nodes in 2026, in San Francisco and Berlin.

Each AI Node in Berlin and SF provides three things:

1. funding for resident researchers, our “AI Nodees”
2. a physical space with residency and programs such as technical workshops and hackathons
3. local compute infrastructure so that researchers working on privacy-preserving and secure AI are not dependent on hyperscaler infrastructure.

In-depth

Our grant funding concentrates on three broadly d/acc aligned research areas:

• AI for better epistemics and forecasting

• security (usable security, access control, automated red-teaming)

• decentralized cooperative AI (local AI, multi-agent systems, governance protocols)

In addition to day to day support in the Node, such as co-working, salons, and research sprints, we host annual field-building workshops to make progress on underexplored AI safety domains. E.g. see our Supercooperation Workshop which happened as an ICML satellite, or our Secure and Sovereign AI Workshop in Berlin.

We provide local compute for our AI Nodees for two reasons: 1) some AI security or privacy-oriented projects need physical access to tinker with hardware or have sensitive computations that should not be on cloud providers, and 2) we believe that extrapolating current compute cost trends and potential government restrictions, it might become uneconomical or simply impossible for independent AI safety projects to get access to sufficient compute long-term. We're trying to hedge against that.

The bigger picture

The physical AI Node locations also diversify where and how AI safety work gets done. The Berlin office addresses a geographic gap: European AI safety capacity is small relative to the US, and research capacity outside a single jurisdiction matters for sovereignty. SF addresses an institutional gap: most work there happens inside frontier labs and a few larger non-profits but our AI Nodes provide an institutional cover for open AI safety work seeking to counter power concentration.

The Nodes are currently funded through the end of 2026. Without further support, the project is unlikely to continue beyond that date. However, we already have substantial interest from groups in London, Boston, and Singapore to start their own Nodes. Long-term, a network of Nodes across jurisdictions sharing compute, talent, and research agendas could function as an independent attractor for AI safety work.