Generative multimodal models can exhibit safety failures that are inherently relational: two benign concepts can become unsafe when linked by a specific action or relation (e.g., child-drinking-wine). Existing unlearning and concept-erasure approaches often target isolated concepts or image-text pairs, which can cause collateral damage to benign uses of the same objects and relations. We propose relationship-aware safety unlearning: a framework that explicitly represents unsafe object-relation-object (O-R-O) tuples and applies targeted parameter-efficient edits (LoRA) to suppress unsafe tuples while preserving object marginals and safe neighboring relations. We include CLIP-based experiments and robustness evaluation under paraphrase, contextual, and out-of-distribution image attacks.
@article{arxiv.2603.14185,
title = {Relationship-Aware Safety Unlearning for Multimodal LLMs},
author = {Vishnu Narayanan Anilkumar and Abhijith Sreesylesh Babu and Trieu Hai Vo and Mohankrishna Kolla and Alexander Cuneo},
journal= {arXiv preprint arXiv:2603.14185},
year = {2026}
}