AIM: Anchor Identity Features, then Match for Multimodal Large Language Model Unlearning

The 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP findings), 2026