Research · Wired AI ·
A New Trick Reveals AI Models’ Inner Thoughts
Researchers developed a method to extract reasoning traces from Claude, GPT, and Gemini models. They report that the traces suggest some Chinese AI systems may have been trained using leading US models.