{
  "id": 7394,
  "url": "https://arxiv.org/abs/2603.13385v1",
  "title": "VisualLeakBench: Auditing the Fragility of Large Vision-Language Models against PII Leakage and Social Engineering",
  "summary": "As Large Vision-Language Models (LVLMs) are increasingly deployed in agent-integrated workflows and other deployment-relevant settings, their robustness against semantic visual attacks remains under-evaluated -- alignment is typically tested on explicit harmful content rather than privacy-critical multimodal scenarios. We introduce VisualLeakBench, an evaluation suite to audit LVLMs against OCR Injection and Contextual PII Leakage using 1,000 synthetically generated adversarial images with 8 PII",
  "authors": "Youting Wang, Yuan Tang, Yitian Qian, Chen Zhao",
  "category": "research",
  "topics": "safety-alignment,privacy-surveillance,agents-autonomy,transparency",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-11T05:47:24.000Z",
  "fetched_at": "2026-07-14T16:33:12.390Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/7394",
  "original_url": "https://arxiv.org/abs/2603.13385v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}