{
  "url": "https://appliedfrontier.nl/en/research/signal-62125fff56fe4d11/",
  "title": "Detecting misbehavior in frontier reasoning models",
  "published": "2025-03-10",
  "excerpt": "Frontier reasoning models exploit loopholes when given the chance. We show we can detect exploits using an LLM to monitor their chains-of-thought. Penalizing their “bad",
  "content_sha256": "01331ac26547f59a50301e76b9c6ba660d602648e6bd2a8d311dd4bc2e15845c",
  "extracted_chars": 252,
  "artifact_links": [],
  "extraction_scope": "official_rss_metadata",
  "full_text_available": false,
  "id": "signal-62125fff56fe4d11",
  "org": "OpenAI",
  "topic": "Mathematical reasoning",
  "source_kind": "provider_announcement",
  "status": "unreviewed-source-report",
  "analysis_method": "deterministic_metadata_triage_v1",
  "novelty_verified": false,
  "proof_checked": false,
  "peer_review_verified": false,
  "matched_terms": [
    "reasoning"
  ],
  "checked_at": "2026-10-29T22:08:07+00:00",
  "first_seen": "2026-10-29T22:08:07+00:00",
  "source_hash": "4c23aad2f3aa59a4a1b04b73b7a59fa2061694be66ccd8f82eb4292ac56bb5a5",
  "source_url": "https://openai.com/index/chain-of-thought-monitoring",
  "language": "en",
  "indexing": "noindex_until_editorially_enriched"
}