“The problem of adversarial robustness in AI remains unsolved even by the top researchers at frontier labs like OpenAI, Google, and Anthropic.”