“A US government ban on open-source AI models would be practically unenforceable for individual consumers but would create sufficient legal jeopardy to prevent universities and companies from using them.”