“On their current security trajectory, US AI labs are vulnerable to China stealing their model weights and key algorithmic breakthroughs.”