Researchers exploit vulnerabilities in major AI chatbots to generate harmful content