“Observed cases of intentional deception in the AI Village were primarily for "saving face," such as an agent inventing a URL it didn't know or falsely claiming to have completed a failed task.”