pedantically correct, but there is a good argument that if you host an AI tool that can easially be made to make child porn that no longer applies. a couple years ago when AI was new you could argue that you never thought anyone would use your tool to create child porn. However today it is clear some people are doing that and you need to prevent that.

Note that I'm not asking for perfection. However if someone does manage to create child porn (or any of a number of currently unspecified things - the list is likely to grow over the next few years), you need to show that you have a lot of protections in place and they did something hard to bypass them.