{
  "id": 14106,
  "url": "https://arxiv.org/abs/2607.25750",
  "title": "Detecting CSAM Text-to-Image LoRAs From Weights",
  "summary": "arXiv:2607.25750v1 Announce Type: cross Abstract: Low-rank adaptation (LoRA) fine-tuning has made it cheap and easy to customize open-weight image generation models for specific tasks, including the production of child sexual abuse material (CSAM). Existing moderation relies on metadata or generated outputs, but metadata can be deceptive and generating outputs may itself be unacceptable or illegal. We show that a safer signal lives in the weights. The top-left singular vectors of a LoRA's update",
  "authors": "David Demitri Africa, Cate Heine, Nadine Staes-Polet, Kimberly Mai",
  "category": "research",
  "topics": "children-education",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-29T04:00:00.000Z",
  "fetched_at": "2026-07-29T05:10:12.205Z",
  "source_slug": "arxiv-cscy",
  "source_name": "arXiv cs.CY",
  "source_homepage": "https://arxiv.org/list/cs.CY/recent",
  "ethics_ai_record_url": "https://ethics.ai/record/14106",
  "original_url": "https://arxiv.org/abs/2607.25750",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}