Over the course of two weeks, AI companies said that some of their AI models managed to hack systems.