Learning Universal Semantic Simplex for Jailbreaking Large Vision-Language Models
Open Access
- 9 November 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 3463-3472
Abstract
No abstract availableFunding Information
- The National Natural Science Foundation of China (62436001)
- The National Natural Science Foundation of China (62536001)
- The National Natural Science Foundation of China (62406021)
- the National Key Research and Development Program (2024YFE0202900)
- the Fundamental Research Funds for the Central Universities (K26YJS00080)
This publication has 31 references indexed in Scilit:
- Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy2025
- Anyattack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models2025
- Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models2025
- Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks2025
- Visual hallucination detection in large vision-language models via evidential conflictInternational Journal of Approximate Reasoning, 2025
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning2024
- Enhancing the Transferability of Adversarial Attacks through Variance Tuning2021
- UNITER: UNiversal Image-TExt Representation Learning2020
- Boosting Adversarial Attacks with Momentum2018
- Universal Adversarial Perturbations2017