Using reinforcement learning to train adversarial agents for automated LLM safety testing - View it on GitHub