{
  "id": 16294,
  "url": "https://arxiv.org/abs/2608.03231v1",
  "title": "Structure-Aware Robust Fine-Tuning: Defending Vision-Language-Action Robots Against Physical Attention Hijacking",
  "summary": "Vision-Language-Action (VLA) policies promise general robotic manipulation, but their robustness against physical-world attacks remains fragile. In particular, we show that physically realizable adversarial patches can reliably induce failures by triggering a mechanism we call policy-critical action-to-vision attention hijacking, where action-conditioned attention is diverted from task-relevant regions to a localized patch. To demonstrate the threat, we propose Attention-Guided Semantic Disrupti",
  "authors": "Jinquan Zhang, Dongfu Yin, Run Yang, Yufeng Yan, Zhen Tian, F. Richard Yu",
  "category": "research",
  "topics": "regulation,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-08-04T07:03:05.000Z",
  "fetched_at": "2026-08-05T05:10:44.550Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/16294",
  "original_url": "https://arxiv.org/abs/2608.03231v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}