arxiv.org web signal

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

By Alexis Dufresne Published September 17, 2026 at 21:12 UTC

Summary

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

Shared on Bluesky by 2 AI experts

Artificial Intelligence Papers: Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection Keertana Chidambaram, Andrew Ilyas, Vasilis Syrgkanis … →
AI Firehose @ai-firehose.column.social: Researchers have revealed a method called "plan injection" that lets adversaries manipulate language models into harmful reasoning through b… →

Originally reported by arxiv.org

Read the original article →

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