A discussion examined how people can remain in control of AI agents, which can make decisions and act independently. It noted that such software agents have existed for decades and that their behavior can sometimes get out of hand. The discussion included the idea that AI could be used to make other AI systems safer.