Can We Trust AI Explanations? Evidence of Systematic Underreporting in Chain-of-Thought Reasoning - Explained Simply | ArXiv Explained