{
  "id": 6867,
  "url": "https://arxiv.org/abs/2603.21687v3",
  "title": "MIRAGE: The Illusion of Visual Understanding",
  "summary": "Multimodal AI systems have achieved remarkable performance across a broad range of real-world tasks, yet the mechanisms underlying visual-language reasoning remain surprisingly poorly understood. We report three findings that challenge prevailing assumptions about how these systems process and integrate visual information. First, Frontier models readily generate detailed image descriptions and elaborate reasoning traces, including pathology-biased clinical findings, for images never provided; we",
  "authors": "Mohammad Asadi, Jack W. O'Sullivan, Fang Cao, Tahoura Nedaee, Kamyar Rajabalifardi, Fei-Fei Li et al.",
  "category": "research",
  "topics": "bias-fairness,healthcare",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-23T08:18:09.000Z",
  "fetched_at": "2026-07-14T16:32:50.143Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6867",
  "original_url": "https://arxiv.org/abs/2603.21687v3",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}