How would you conduct adversarial testing to uncover jailbreaks in a conversational AI agent at Character.AI?
