ニュース

"Paraphrasing attacks" could modify text so an AI misclassifies. Meanwhile, it's meaning would remain unchanged to humans.