← back
arXivYuxuan Huang, Xingyu Zeng, Tianhang Zheng, Chaochao LuWed, Aug 5, 2026, 9:55 AM PDT
score 17.2

New AI defense blocks harmful fine-tuning without limiting useful training

Original: Gradient Immunity: Null-Space Resistance to Malicious Fine-Tuning

Source: arxiv.org

Writing ELI5 summary…