Jailbreaking techniques for AI models are shared among networked adversaries like terrorist groups.