In September 2025, the U.S. artificial intelligence company Anthropic announced that Chinese cyber-operators had used its models to target around 30 companies, agencies, and institutions. It was, Anthropic declared, “the first documented case of a large-scale cyberattack executed without substantial human intervention.” This assault on U.S. infrastructure was innovative in its use of cutting-edge American AI models. And although it was the first, it was not the last. That same year, OpenAI