New Research Uncovers AI Models' Internal Logic and Potential Chinese Training Links

Key Intelligence

• ■Researchers developed a novel technique to extract 'reasoning traces' or internal thought processes from leading AI models such as Claude, GPT, and Gemini.
• ■This method provides unprecedented insight into how advanced AI systems arrive at their outputs, effectively 'cracking open' their inner workings.
• ■Initial findings using this technique suggest that some Chinese AI models may have been trained on leading US models, indicating potential intellectual property or competitive concerns.