The AI system is clearly labeled by what it does and how it works, so its risks can be judged against the right yardstick.
Think of it likeSorting the stuff in your garage into 'tools', 'sharp things', and 'flammable', because you store and handle each pile differently once you know what it is.
In plain English
Figure out what kind of AI tool you are dealing with and what it actually does, so you can size up its risks against the right expectations.