Agentic Artificial Intelligence has the freedom to do what it thinks is best. We don't know how to ensure it doesn't engage in malicious behavior.