OpenMined builds open-source infrastructure that lets AI systems use data without anyone having to hand it over. Our research asks what becomes possible when that infrastructure exists.
Who gets to know what? Who gets paid? Who gets to say no?
We work across 3 primary pillars.
Intelligence composed live over a network, rather than bundling all data, compute, and algorithms into a single centralized source owned by one entity. A conventional AI system is a trained model plus, optionally, a RAG layer over data that happens to be colocated or on the internet. NSAI decentralizes both halves. The model context can come from many independent data owners, and the model itself can be an ensemble of models from many independent owners.
A relationship between sources of AI-related data, compute, and talent and the AI users receiving predictions. These two groups each have ABC when resource owners dynamically control how much intelligence to give each AI prediction they support, and AI users dynamically control which resource providers they want to rely on for each AI prediction.
A framework for governing information flows so that parties can collaborate on data without giving up control of it. It names five guarantees a flow can need: input privacy, output privacy, input verification, output verification, and flow governance. Privacy-enhancing technologies (PETs) are the mechanisms that supply those guarantees, and ST provides a vocabulary for saying which ones a given flow is missing.
OpenMined Research focuses on 5 distinct areas for deeper exploration.
PhD students and predoctoral researchers who join part-time alongside their academic work, with the space, resources, and community for ambitious research. Meet the inaugural 2026 cohort.
Compute, engineering, and evaluation for researchers bringing ensemble and routing methods to today’s frontier models, with a stipend for your time.