The problem is that it is very difficult to automate picture censorship re: inappropriate images.
As humans, with a quick glance we can determine inappropriate pictures.
For a computer program - even with AI it's not that simple (it's very complex).
For example think of a picture of an adult in perfectly appropriate swimwear.
What is the colour of skin?
What "parts" are appropriately covered?
How do you define to a computer "appropriate parts"?
Especially when you consider the variation in human bodies and then combine with the angle the picture was as taken, the lighting, the distance (close up or far away) or how a picture is cropped.
There is no foolproof technology as stands.
There is tech that can filter, but it won't catch everything.
MNHQ may be a target right now, but other (massively bigger) platforms are subject to this issue.