I think it’s just an easier way for the news to describe it to the every man. In reality, this is a combination of sandbox failures, what I’d have to imagine is incorrect prompting, and most concerningly, alignment issues in the models themselves.