PolyJailbreak: Cross-Modal Jailbreaking Attacks on Black-Box Multimodal LLMs
cs.CR
Submitted: 2025-10-20
Updated: 2026-03-07
DOI: 10.1109/TDSC.2026.3707228
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Gemini: A Family of Highly Capable Multimodal Models
- Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
- DeepSeek-V3 Technical Report
- The Llama 3 Herd of Models
- GPT-4o System Card
- FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space
- FlipAttack: Jailbreak LLMs via Flipping
- UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
- X-Fusion: Introducing New Modality to Frozen Large Language Models
- RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering
- The Dark Side of Trust: Authority Citation-Driven Jailbreak Attacks on Large Language Models
- Mind the Inconspicuous: Revealing the Hidden Weakness in Aligned LLMs' Refusal Boundaries
- HyperLLaVA: Dynamic Visual and Language Expert Tuning for Multimodal Large Language Models
- Jailbreaking Multimodal Large Language Models via Shuffle Inconsistency
- Universal and Transferable Adversarial Attacks on Aligned Language Models
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills
- Token Composition: A Graph Based on EVM Logs