Intelligence should be public infrastructure — open, decentralized, and accessible to all.

The foundational technologies of artificial intelligence — models, datasets, evaluations, agent protocols, and the standards that bind them — are becoming the infrastructure of everything else. Infrastructure that important should be open, decentralized, and accessible. It should belong to the public, not to the few.

Today it is being enclosed. Compute is concentrated in a handful of clouds and export-controlled at national borders. Frontier models are trained on the world’s data — largely without its consent — and served back to the world on revocable terms. Individuals and businesses now route their daily work, and their trade secrets, through AI systems they hold no rights in: access can be withdrawn, repriced, or restricted at any time, by a provider’s decision or a government’s.

This is not an argument against companies, or against oversight. It is an argument about defaults. The web faced the same fork and chose openness — standards nobody owned, infrastructure anyone could join — and a generation built on it without asking permission. The foundation’s work is to make that choice available for AI: to design open infrastructure, to develop the standards that make an open agentic web possible, and to hold AI systems accountable and transparent — so that as autonomous agents take on critical decisions, those decisions remain under meaningful oversight.

Other organizations advance pieces of this work — publishing research, releasing weights, advocating policy, shepherding open source. The foundation brings three of those pieces together — research, governance, and standards — so that the way AI is built, evaluated, and adopted is open in its methods, accountable in its consequences, and inclusive in its reach.