We're doing that internally to continuously improve our own agent and make it robust against adversarial attacks itself. We will release some insights about self-improvement soon!

Cool, sounds like an interesting experiment!