{
  "id": 6233,
  "url": "https://arxiv.org/abs/2604.06377v3",
  "title": "The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment",
  "summary": "We investigate whether post-trained capabilities can be transferred across models without retraining, with a focus on transfer across different model scales. We propose the Master Key Hypothesis, which states that model capabilities correspond to directions in a low-dimensional latent subspace that induce specific behaviors and are transferable across models through linear alignment. Based on this hypothesis, we introduce UNLOCK, a training-free and label-free framework that extracts a capabilit",
  "authors": "Rishab Balasubramanian, Pin-Jie Lin, Rituraj Sharma, Anjie Fang, Fardin Abdi, Viktor Rozgic et al.",
  "category": "research",
  "topics": "safety-alignment,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-07T19:02:10.000Z",
  "fetched_at": "2026-07-14T16:32:20.056Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6233",
  "original_url": "https://arxiv.org/abs/2604.06377v3",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}