No AI summary available for this article.
Why It Matters
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.
Provenance
Discovered via ArXiv and published by ArXiv.
Key Claims
Original description
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 "plan injection". We initially discover this attack in the multiple-choice question-answering monitorability setting proposed by Lanham et al. (2023), using the investigator-agent elicitation framework of...
Discovered via ArXiv
Research papers and preprints from arXiv.
Publisher: arxiv.org
ID: http://arxiv.org/abs/2609.15989v1 · Indexed about 1 hour ago