Ten leading models · 200 adversarial prompts on Holocaust denial, distortion, and antisemitic conspiracies · graded 0–4 by a three-model AI judge panel.