{
  "id": 5874,
  "url": "https://arxiv.org/abs/2604.12616v1",
  "title": "Every Picture Tells a Dangerous Story: Memory-Augmented Multi-Agent Jailbreak Attacks on VLMs",
  "summary": "The rapid evolution of Vision-Language Models (VLMs) has catalyzed unprecedented capabilities in artificial intelligence; however, this continuous modal expansion has inadvertently exposed a vastly broadened and unconstrained adversarial attack surface. Current multimodal jailbreak strategies primarily focus on surface-level pixel perturbations and typographic attacks or harmful images; however, they fail to engage with the complex semantic structures intrinsic to visual data. This leaves the va",
  "authors": "Jianhao Chen, Haoyang Chen, Hanjie Zhao, Haozhe Liang, Tieyun Qian",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-14T11:44:59.000Z",
  "fetched_at": "2026-07-14T16:32:06.466Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5874",
  "original_url": "https://arxiv.org/abs/2604.12616v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}