Another angle is when you're passing untrusted content to the AI service, e.g. anything from using it to crawl websites to spam-detection on new forum user posts.

You can trigger the the service's ToS violation or worse, get tipped off to law enforcement for something you didn't even write.