there needs to be a difference between vibe-coded and developed using AI. When building complex systems, you would be aware that building the systems and then coding it takes a lot of time, so im not architecting it using AI, im architecting it myself, but developing it using AI, there has to be a clear difference between both of them.
Im with you on the fact that on the dark web there is dark stuff, but that dark stuff comes in through images, urls or some sort of media right? What if all media itself is blocked, thats exactly what ive done, and then whatever comes through text will still needs filtering, that gets handled in multiple levels, it goes through a clear wordlist, which kindof filters it out, and then a classifier and then, if AI is enabled, the AI handles anything that might be left off, but never comes to that.
Also AI is not needed to run this tool, it just helps out in making decisions, the engine itself does its job carefully, that even without AI it gives very close results, but the only thing it kindof stutters on is when there are pivots, and thats currently beign worked on.
On the part of that NSFW material itself being the intel needed, agreed but to some extinct, 90% of what the ethical analysts are trying to find is something realted to their own domain being sold, or database pertaining to them, so for that case this does a pretty good job and for that 10% there is not automation and the analysts are trained enough to know what to exactly. im not trying to replace the actual analysts but to accelerate their job basically.
Hopefully you understand that AI is just an addition and the engine itself works well without any AI whatsoever, but yeah on the part of NSFW filtering, i would stand with it, that this tool is capable of filtering anything that gets thrown at it, mainly because 80-90% nsfw stuff is media, which is already blocked.