Securing AI Model Weights: Preventing Theft and Misuse of Frontier Models

The goal of this report is to improve the security of frontier artificial intelligence (AI) or machine learning (ML) models. (Frontier models are those that match or exceed the capabilities of the most advanced AI models at the time of their development.) Our analysis focuses on foundation models, and specifically large language models and similar multimodal models. We focus on the critical leverage point that is the core of a models intelligence and capabilities: its weights, a term used here to refer to all learnable parameters derived by training the model on massive datasets. These parameters stem from large investments in data, algorithms, compute (i.e., the processing power and resources used to process data and run calculations), and other resources; compromising the weights would give an attacker direct access to the crown jewels of an AI developers work and the ability to exploit them for their own use.