[ICLR'26 Oral] Beyond Prompt-Induced Lies: Investigating LLM Deception on Benign Prompts - View it on GitHub