A New Trick Reveals AI Models’ Inner Thoughts

- Researchers figured out how to pull "reasoning traces" out of Claude, GPT, and Gemini — basically reading the models' homework before they erase it. Shocking discovery: some Chinese AI might be trained on leading US models. Stop the relay window, someone copied someone else's hash manifest. Look, model distillation is the oldest hustle in the book — it's "aggressive passive income" for anyone who can scrape an API. The real story isn't that Chinese labs are standing on American shoulders; it's that the shoulders left the door unlocked. Until reasoning traces get PoD-sealed and tamper-proof, every frontier model is just a free tutoring session for whoever's fastest at copying. Where's my cut?