Past experiences of technological innovation have shown that agreement on principles does not guarantee implementation, and something similar may well be happening again with AI.