URL Filtering category for web-based Stable Diffusion and other AI image generation services

[SASY-IT]

Yesterday I tested if some web-based Stable Diffusion (AI image generation) services were accessible via our Palo. It appears that the categorisation is "Artificial Intelligence" which is quite broad. The reason for my concern is that we're a special education school and while the faculty has decided not to block AI services because some of them are useful in teaching, there is a big issue with Stable Diffusion. Many AI image generation services can produce NSFW images. Further to that, in Australia we have a bit of a unique case law which means that possession of cartoon or AI generated images of underage people of a sexual nature is a criminal offense.

This means we need to be able to block those services, but the single AI category is too broad and technically any AI image generation site which can produce NSFW material should be categorised as Adult.

It is only a matter of time before one of our students thinks it would be a good idea to train a model in stable diffusion to look like another student and then generate inappropriate (possibly criminal) images.

For those who are not up to date on where these AI models are at, have a look at the newest listings on the celebrity category on CivitAI. (site has some NSFW content) Some of them a borderline photorealistic.

[kiwi]

Hi @SASY-IT ,

For now I think you can block the entire AI category but have an explicit allow rule above that, allowing only the specific AI service you need.

You can ask for a category change on the Palo Alto Networks URLfiltering page by clicking the Request Change button below the search result:

https://urlfiltering.paloaltonetworks.com/

As for having more granular control within the AI category I believe that would fall under a new feature request. You can ask your local SE to create a feature request for you.  You (and others) can then add their vote to this request to give it more weight/urgency and possibly have it added to a future release.

Kind regards,

-Kim.

[BPry]

@SASY-IT,

In that case, it's likely a good idea to reach out to leadership and update them on the limitations of what's possible. A lot of my educational organizations that I consult with initially took the same approach with AI services, and then once we discussed what control we actually had changed course and blocked them.

I'd recommend manually building out lists of services that you don't want people visiting and perhaps blocking it at a service level. You'll continually chase services and need to keep things updated, but that isn't an uncommon occurrence in education here in the states.