arXiv cs.AI / cs.LG / cs.CL·12d agoCorrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection#ai-safety#chain-of-thought-monitoring#deceptionAI safety & security