{
  "id": 7486876,
  "title": "Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection",
  "url": "https://urgent.news/2026/09/14/corrupt-plans-clean-traces-evading-chain-of-thought-monitoring-with",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-14T18:33:27.000Z",
  "source": {
    "name": "arXiv cs.AI",
    "slug": "arxiv-cs-ai",
    "url": "https://arxiv.org/abs/2609.15989v1"
  },
  "original_language": "en",
  "account": null,
  "summary": "Chain-of-thought (CoT) monitoring is a safety strategy where the reasoning of a large language model \"actor\" is inspected by a \"monitor\" (often another language model) for signs of unsafe planning, deception, or misalignment. We find that planting harmful but benign-sounding reasoning in the actor's context can steer it to perform adversarial actions while evading monitors, an attack we term…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}