Chinese AI labs are believed to be ahead of the AI safety community in developing techniques for extracting reasoning traces from closed models