The authors describe how to secure the weights of frontier artificial intelligence and machine learning models (that is, models that match or exceed the capabilities of the most advanced models at the time of their development).