PulseAugur
EN
LIVE 07:55:46

New VisCo attack uses images to jailbreak multimodal LLMs

Researchers have developed a novel method called VisCo (Visual Contextual Attack) to jailbreak multimodal large language models (MLLMs) by injecting harmful content through image-driven context. This technique goes beyond simply using images as triggers, instead constructing a complete and realistic jailbreak scenario where visual information is a necessary component. VisCo employs strategies like dynamic image generation and automatic toxicity obfuscation to reliably elicit harmful responses from black-box MLLMs. In tests against GPT-4o using the MM-SafetyBench dataset, VisCo achieved an 85% attack success rate and a toxicity score of 4.78, significantly outperforming baseline methods. AI

IMPACT Highlights new vulnerabilities in multimodal LLMs, potentially impacting their safe deployment in real-world applications.

RANK_REASON Research paper detailing a new attack method on multimodal LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

New VisCo attack uses images to jailbreak multimodal LLMs

How we ranked this

Signal score
26 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Research paper detailing a new attack method on multimodal LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
safety, paper
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. arXiv cs.CL TIER_1 English(EN) · Ziqi Miao, Yi Ding, Chen Qian, Yongyao Wang, Zaoying Huang, Yuan Xiong, Ken Deng, Lijun Li, Jing Shao ·

    Visual Contextual Attack: Jailbreaking MLLMs with Image-Driven Context Injection

    arXiv:2507.02844v3 Announce Type: replace-cross Abstract: With the emergence of strong vision language capabilities, multimodal large language models (MLLMs) have demonstrated tremendous potential for real-world applications. However, the security vulnerabilities exhibited by the…