“Some AI research suggests that the "chain of thought" reasoning traces displayed by models may not be a truthful representation of their internal process, but rather a fabricated output designed to mimic human reasoning.”