{
  "id": 4848,
  "url": "https://arxiv.org/abs/2605.08245v4",
  "title": "When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models",
  "summary": "Vision-Language Models (VLMs) increasingly power high-stakes applications, from medical imaging to autonomous systems, yet they routinely hallucinate, confidently describing content not present in the input. We investigate the root causes of these failure modes with a mechanistic analysis focusing on the decoder-based VLMs. We trace these failure modes to a geometric over-alignment: to bridge the modality gap required by attention mechanisms, decoder-based VLMs over-align visual embeddings with ",
  "authors": "Harshvardhan Saini, Samyak Jha, Yiming Tang, Dianbo Liu",
  "category": "research",
  "topics": "safety-alignment,healthcare,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-07T10:09:18.000Z",
  "fetched_at": "2026-07-14T16:31:17.583Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4848",
  "original_url": "https://arxiv.org/abs/2605.08245v4",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}