{
  "id": "2606.25013",
  "title": "Do Thinking Tokens Help with Safety?",
  "authors": "Narutatsu Ri, Abhishek Panigrahi, Sanjeev Arora",
  "first_public": "2026-06-23",
  "added_at": "2026-11-02",
  "venue": "arXiv",
  "arxiv": "2606.25013",
  "paper_url": "https://arxiv.org/abs/2606.25013",
  "html_url": "https://arxiv.org/html/2606.25013v1",
  "lens": [
    "analysis"
  ],
  "representation": [
    "hidden-state"
  ],
  "computation": [
    "token-wise"
  ],
  "training": [
    "analysis-only"
  ],
  "questions": [
    "interpretability",
    "evaluation",
    "limitations"
  ],
  "evidence": [
    "empirical",
    "probing",
    "boundary"
  ],
  "modality": [
    "language"
  ],
  "summary": "Probes whether the first hidden state of a reasoning trace predicts a model’s eventual refusal or compliance. Compares this early signal with partial-trace continuations and thinking-enabled versus direct responses to test whether written deliberation changes safety decisions.",
  "provenance": "Primary-source literature review, 2026-11-02",
  "author_names": [
    "Narutatsu Ri",
    "Abhishek Panigrahi",
    "Sanjeev Arora"
  ],
  "record_url": "https://latent-implicit-thinking.github.io/papers/2606.25013/",
  "legacy_record_url": "https://latent-implicit-thinking.github.io/papers/#paper-2606-25013",
  "citation_date": "2026-06-23",
  "json_url": "https://latent-implicit-thinking.github.io/papers/2606.25013/record.json",
  "markdown_url": "https://latent-implicit-thinking.github.io/papers/2606.25013/index.md",
  "bibtex_url": "https://latent-implicit-thinking.github.io/papers/2606.25013/citation.bib",
  "csl_url": "https://latent-implicit-thinking.github.io/papers/2606.25013/citation.csl.json",
  "pdf_url": "https://arxiv.org/pdf/2606.25013"
}