Something disturbing happened with an AI model Anthropic researchers were tinkering with: it started performing a wide range of “evil” actions, ranging from lying to telling a user that bleach is safe ...