cdt comments on Disempowerment patterns in real-world AI usage