o1 (like all LLMs) is like the horse Clever Hans, it picks up subtle cues to guess the answer from the questioner who already knows the answer. If you ask o1 "how do you know that ___?" it answers X, if you ask it "are you sure that ___?" it answers ¬X.

?