Source · DiggingBeagle record
Image-based Prompt Injection: Hijacking Multimodal LLMs through Visually Embedded Adversarial Instructions
Each support, contradiction or context label applies to a cited Claim, not to a whole Case.
Source record
Claim-level citations (2)
- supportsStealthy image-embedded instructions manipulated multimodal LLM behavior: The authors report attack success up to 64% for their most effective tested configuration under stealth constraints.
Abstract, reported success rate
- supportsStealthy image-embedded instructions manipulated multimodal LLM behavior: The image-based prompt-injection pipeline hides adversarial instructions inside natural images using region selection, adaptive text scaling and background-aware rendering so the multimodal model can interpret the instruction.
Abstract
Cite this record
DiggingBeagle. “Image-based Prompt Injection: Hijacking Multimodal LLMs through Visually Embedded Adversarial Instructions.” https://diggingbeagle.com/sources/image-based-prompt-injection-hijacking-multimodal-llms-through-visually-embedded/
Citation guidance