Signals 4 · free daily AI digest

Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection

Paper recorded by Signals 4 on 2026-09-14 in cs.AI. Abstract reproduced from arXiv; link to the original below.

Published 2026-09-14 on arXiv · recorded by Signals 4 on 2026-09-15

Category: cs.AI · 人工智能 · first seen 2026-09-15

Abstract

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

Read on arXiv →

#61 most recent of 300 cs.AI papers we have recorded · ↑ newer: Self-Emergence Agent Architecture:Behavior-Inertia HMM, Reflexive Meta · ↓ older: Stellar Colosseum: A Many-Agent Harness for Long-Horizon Research in M
Cite this page: Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection: the #61 most recent of 300 cs.AI papers we have recorded (as of 2026-09-14). Source: Signals 4 (Signals API) — https://data.jiangzhang.ca/signals4/t/papers/corrupt-plans-clean-traces-evading-chain-of-thought-monitoring-with-plan-injecti.html
Free to quote with attribution to “Signals 4 (Signals API)”. Machine-readable: papers.json
Related: More cs.AI papers · arXiv signals · All papers · Today in AI
Get 4 AI signals a day by email — free.
Subscribe free → See all plans →
Get 4 AI signals a day by email — free
All models · All repos · By company · Daily editions