Researchers have discovered OpenAI's GPT-4’s security settings are ill-equipped to handle languages not commonly fed into its training data.
