What do AI firms do when users tell chatbots their dark, violent thoughts?

What do AI firms do when users tell chatbots their dark, violent thoughts?