{
  "id": 14189,
  "url": "https://blog.redwoodresearch.org/p/untrusted-advice-for-ai-control-short",
  "title": "Untrusted advice for AI control: Short, strong advice significantly uplifts weak LLMs",
  "summary": "When a misaligned AI can only output tiny amounts of information, it may find sabotage very difficult",
  "authors": "Caleb Biddulph",
  "category": "org",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-27T23:59:30.000Z",
  "fetched_at": "2026-07-29T05:10:12.205Z",
  "source_slug": "redwood",
  "source_name": "Redwood Research",
  "source_homepage": "https://blog.redwoodresearch.org",
  "ethics_ai_record_url": "https://ethics.ai/record/14189",
  "original_url": "https://blog.redwoodresearch.org/p/untrusted-advice-for-ai-control-short",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}