AI model weights are the billions or trillions of numerical parameters that result from training a machine learning model on massive datasets using enormous computing power. A sufficiently powerful set of weights represents a distillation of vast computational resources — downloading them gives the user access to AI capabilities equivalent to those that required frontier hardware to produce. Biden's AI Diffusion Rule was the first major attempt to include model weights alongside physical chips in export controls, recognizing that software-only distribution of powerful AI could circumvent hardware restrictions. The practical enforcement challenge is significant: model weights are data that can be transmitted digitally, making them harder to control than physical chips.