This would be a fairly insane breach of trust and common sense if true; the chain-of-thought / reasoning trace is, from an information perspective, close to a superset of the prompt and model response.