Can AI predict and preemptively strike adversarial ai development before it becomes operational?