These newer models appear more likely to indulge in rule-bending behaviors than previous generations—and there’s no way to stop them. Facing defeat in chess, the latest generation of AI reasoning ...