Researchers Extract Hidden Reasoning Traces From Major AI Models

A team of researchers has developed a technique to extract so-called reasoning traces from leading AI models, including Claude, GPT, and Gemini. These traces reveal the internal thought processes that models use when generating responses. The findings suggest that certain Chinese AI models may have been trained using outputs or data from top US-developed models. The discovery raises significant questions about intellectual property and the integrity of AI development practices. Researchers say the method offers a new window into understanding how large language models actually operate.