Encrypted chain-of-thought blocks returned by Anthropic, OpenAI and Google APIs—we exploit this to decode hidden reasoning at scale.