Honestly, it's not a terrible idea to start asking people to run AI agents as a red team to find vulnerabilities.