Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness Weilin Lin 1 Li Liu 1 Shaokui Wei 2 Jianze Li3,4,2

Nov-17-2025, 06:36:47 GMT–Neural Information Processing Systems 

However, the behavior of clean unlearning is still under-explored, and vanilla fine-tuning unintentionally induces back the backdoor effect.