LeCun makes the political case for open-source AI as the only path to sovereignty

At the UN’s Open Source Week, Yann LeCun—Meta’s former chief AI scientist—argued that open-source models are the only viable route to AI sovereignty for most of the world. He frames AI as coming infrastructure that will soon mediate nearly all interaction with digital information, and warns that letting a handful of proprietary systems from US and Chinese tech giants control that layer threatens linguistic diversity, democracy, and human rights. Most countries can’t afford to train frontier models alone, he contends, but a shared open platform built collaboratively could eventually surpass closed systems.

His proposed mechanism is federated rather than centralized. Through Project Tapestry—a bottom-up, open-signup effort he launched after leaving Meta alongside the AI Alliance—each country or institution would digitize its own cultural material and contribute to a global model by exchanging parameter vectors instead of raw data, preserving local sovereignty. LeCun cites early interest from European nations, the UK, UAE, India, Japan, Korea and others, plus IBM, NVIDIA, AMD and Intel, and hopes for production by early 2027. He leans on a historical analogy: just as commodity hardware and open-source software wiped out proprietary stacks from Sun, Dell and HP in the early 2000s, he calls the same shift in AI “inevitable.”

Much of the keynote was a rebuttal to safety-driven arguments for restricting open models, which LeCun dismisses as “medieval obscurantism” akin to limiting the printing press. He argues existential and security risks are wildly overstated—a bioweapon recipe isn’t the real bottleneck, and AI that finds vulnerabilities can equally harden defenses—and that the greater danger is using speculative threats to lock the technology inside a few corporate and geopolitical silos. Echoing his concern about concentration, he insists no AI system is unbiased, so societies need a diversity of AI assistants for the same reason they need a diverse press. Delegates from Morocco, Sierra Leone, Jamaica and Spain’s AI regulator voiced agreement.