The research paper titled 'Stealing Reasoning Traces from Proprietary LLMs' demonstrated techniques for extracting reasoning traces from closed models