EviGuard: Evidence-Guided Connector Intervention for Cross-Modal Safety Unlearning in Multimodal LLMs
Open Access
- 9 November 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 3502-3511
Abstract
No abstract availableFunding Information
- New Generation Artificial Intelligence-National Science and Technology Major Project (2025ZD0123704)
This publication has 34 references indexed in Scilit:
- SAUCE: Selective Concept Unlearning in Vision-Language Models with Sparse Autoencoders2025
- Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy2025
- FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual PromptsProceedings of the AAAI Conference on Artificial Intelligence, 2025
- Rethinking machine unlearning for large language modelsNature Machine Intelligence, 2025
- MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models2024
- MLLM-Protector: Ensuring MLLM’s Safety without Hurting Performance2024
- Erasing Concepts from Diffusion Models2023
- G-Eval: NLG Evaluation using Gpt-4 with Better Human Alignment2023
- Knowledge Unlearning for Mitigating Privacy Risks in Language Models2023
- Overcoming catastrophic forgetting in neural networksProceedings of the National Academy of Sciences, 2017