The critical point actually crossed is therefore not “Can AI train its own successors?”, but “Will AI still play by the rules of the exam humans set?”