Testing a natural language model using curiosity driven red teaming prevents LLMs from producing toxic content.