Apple Researchers Propose Locking Pretrained Weights to Block Unauthorized Fine-Tuning

Apple researchers propose a technique to 'lock' open-weight model parameters so they resist unauthorized fine-tuning while still allowing legitimate adaptation.