“Human-in-the-loop security models are ineffective for defending against AI-generated cyber threats due to speed disparities.”