The lack of guardrails leaves AI chatbots open to manipulation and has resulted in them generating antisemitic and other hateful online content, researchers said.