Ars Technica: How often do AI chatbots lead users down a harmful path?. “While the results show that these kinds of manipulative patterns are relatively rare as a percentage of all AI conversations, they still represent a potentially large problem on an absolute basis.”