I posted the other day that you can clean up your object storage from CSAM using my AI-based tool. Many people expressed the wish to use it on their local file storage-based pict-rs. So I’ve just extended its functionality to allow exactly that.
The new lemmy_safety_local_storage.py will go through your pict-rs volume in the filesystem and scan each image for CSAM, and delete it. The requirements are
- A linux account with read-write access to the volume files
- A private key authentication for that account
As my main instance is using object storage, my testing is limited to my dev instance, and there it all looks OK to me. But do run it with --dry_run if you’re worried. You can delete lemmy_safety.db and rerun to enforce the delete after (method to utilize the --dry_run results coming soon)
PS: if you were using the object storage cleanup, that script has been renamed to lemmy_safety_object_storage.py
I hope people share the positive hits of CSAM and see how widespread the problem is…
DRAMTIC EDIT: the records lemmy_safety_local_storage.py identifies, not the images! @bamboo@lemmy.blahaj.zone seems to think it “sounds like” I am ACTIVELY encouraging the spreading of child pornography images… NO! I mean audit files, such as timestamps, the account that uploaded, etc. Once you have the timestamp, the nginx logs from a lemmy server should help identify the IP address.
What a hilarious mistake haha
It is not even a mistake, it’s some pretty mind-fucked up on part of @bamboo@lemmy.blahaj.zone to jump to such a conclusion. crap
It’s cool, most everybody knows what you mean lol. Glad you clarified so there wouldn’t be future misunderstandings
Oh come on, its just a correction of communication
It’s probably projection. Nobody reasonable would have jumped to the same conclusion. It doesn’t even remotely read like that.
It sounds to me like an NT desire for perfectly crafted arguments, without ambiguity. I do this, and feel fortunate that I didnt call for a correction, myself. See how vicious you all are about it.
Oh come on. Being ND doesn’t mean your mind jumps to sharing child porn. That’s a fuckin cop out.
You are saying the mind jumps, but that is the topic. I meant to say that being ND can create a desire for clarity in communication. A direct or terse argument.
I hope people share the positive hits of CSAM and see how widespread the problem is…
It sounds like you’re encouraging people to share CSAM images found, which is obviously not the intent of this tool. There’s probably a better way to phrase what you were trying to say.
I expect they just want statistics
Yes. odd how people think sharing CSAM is why people would post here, instead of actually tracking down and prosecuting those sharing CSAM. Details about the users who sharedl CSAM content, such as timestamps - would help identify the offenders for prosecution.
I assumed as much after reading it several times, but just wanted let you know and point out that statement could be misconstrued. Thanks for clarifying!
Because of the way pictrs organizes photos, which I believe is by hash (could be random id but I suspect not), you should be able to share filenames for cleanup by neighbors without having to share the contents.
Even if it’s not organized that way automatically, though, you can pretty easily use
sha256sumto get a shareable hash before deleting the content.I think timestamps of files would be one of the easier things, and try to track back to postings and comments that references the upload… ideally the logged-in account (which is the standard install of lemmy, only logged-in users can upload to pictrs)
I think both are equally important. I think there’s potential for a centralised database of the metadata involved here. There is likely just one person, or a very small group, committing this current attack. It’s very likely a troll that’s gotten a small handful of images from some dreadful shock site, rather than someone who knows how to access this content more broadly. It might be possible to entirely block the ability of new accounts to keep posting those images using the hashes.
It sounds like you’re encouraging people to share CSAM images found, which is obviously not the intent of this tool.
Yes, that is in fact the context.
Context: "which is obviously not the intent of this tool. "
it is not my intent to share the images, nor is it the context of the tool… Sharing details about the users, timestamps - would be the obvious context.
Why are you such a weirdo where that’s where your mind goes.
Sharing positive hits isn’t saying share the images. It’s saying share the data on who what when where how the hit showed up positive.
Who shared it.
What was it (this is obviously going to be some kind of CSAM given that’s the tool).
When did they share it (time stamps).
Where did they share it (was the same image hit on other runs and what instances did it hit on with the tool).
How did they do it (local sharing, an image hosting service, etc).
You list great points of information that should be shared to admins of other instances. Thanks for clarifying.
In order to get the answers you’re looking for, put out the question “what exactly does this statement mean?” instead of “sounds like this means ___” and waiting for a confirmation/rejection of your assumption.
Removed by mod
We Evolve!
There something really satisfying about witnessing a community starting to talk about serious a issue and days later see things already improved.
Lemmy is now the internet, glory to all volunteer devs. Lets make it the best place we possibly can!
I’m curious… How does one even test such a thing before distributing it without having offending files to test against.
Like during the development process of this project, how on earth can you test it properly? 😂
it uses a model that describes a photo, then it searches the generated description for some terms and ranks the image to some levels of safety.
to test it you use a more general filter, for all nsfw for example, and see if the matches are correct.
Hmm, thinking out loud… Wouldn’t that also make it easy to remove scat porn and Hitler + Nazi flag images?
There has been a lot of spam like that on Lemmy and at least the latter is somewhat illegal to host in Germany as well.
yes… maybe.
as the dev said, it flags a lot of false positive. so a human should look at them anyway.
maybe when this is a bit more evolved, we can use it to preprocess posts, and if a post gets flagged for something, a mod / admin needs to approve the post manually.
maybe for CASM, it gets sent to an external service specialized to that stuff, so the mod / admin doesn’t have to look at the images.
While it’s not the case for this project I’m sure there’s some poor researcher our there who trained a model on actual confiscated CSAM. Or most likely overworked traumatized thirld world content moderators employed by the likes of Meta.
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
Fewer Letters More Letters HTTP Hypertext Transfer Protocol, the Web IP Internet Protocol SSH Secure Shell for remote terminal access nginx Popular HTTP server
3 acronyms in this thread; the most compressed thread commented on today has 8 acronyms.
[Thread #93 for this sub, first seen 31st Aug 2023, 04:15] [FAQ] [Full list] [Contact] [Source code]
This is fantastic work to an immediate problem. Thank you.
There is something amazing about someone just sharing a solution like this without expectation of anything back, and even if this isn’t the best right solution, it contributes to the global commons, and improves society.
It’s great that there’s now a tool, but this kind of issue is why I’m not considering self-hosting a fediverse service: due to the nature of the beast even a single-user instance effectively becomes a publicly accessible distributor of content that others created/uploaded. I’m sure this could be restricted somewhat (by making the web UI inaccessible for the public), but the federation means that other instances need to be able to get content from the server. That’s way to much legal risk for me.
Actually the other instances don’t pull content. You push it out instead. Basically if you don’t run pict-rs and you don’t allow user registreations, you’re safe for everything except potential copyright infringement from some text someone might post. If you make your webui inaccessible to anyone but your IP, you protect against even that.
what stops other instances from pushing bad stuff to you? even if someone I follow posts illegal stuff, it’ll end up on my server.
If you don’t run pictrs, they literally cannot push images to you as you can’t store them
Thanks for adding this. I guess I now have my weekend planned for moving Pict-rs to a server with a fast enough GPU and try this out 🤔
You don’t need to move pict-rs to a GPU server. In fact that would be prohibitedly expensive long-term. I suggest you just use your PC to run this against your current pict-rs server, or just rent a GPU server for this time.
I have a server with a smaller Nvidia GPU available. I hope the 3GB vRAM it has will be sufficient.
OK, but just to point out that the script hasn’t been setup to run locally yet. Just through ssh. Making it look for the files locally is my next to-do
Ah, I was already wondering why it would need SSH keys to scan local files 😅 Well, I only have time to look into this on Sunday I think, so no rush.
You could always SSH into localhost I assume.
Yes that would work, but it’s a bit unnecessary
Yeah, just meant it should be a valid workaround in the event they want to try it before your local scan feature is ready.
seems like clip-interrogator isn’t explicitly supported on apple silicon, I’ll give it a try later
thanks for making this!
Something that might be useful long term is trying to train an AI and release weights to identify CSAM that admins can use to check images. The main problem is finding a way to do this without storing those kinds of images or video :/
My understanding is that right now, the main mechanisms involved use several central databases which use perceptual hashes of known CSAM material. The problem is that this ends up being a whackamole solution, and at least in theory governments could use these databases to censor copyrighted or more general “unapproved” content, though i imagine such a db would lose trust quickly and I’m not aware of this being an issue in practise.
One potential solution is “opportunistic training” where, when new CSAM material gets identified and submitted to the FBI or these databases by various server admins, a small amount of training is done on the AI weights before the image or video is deleted and only a perceptual hash remains. Furthermore, if a picture is reported as “known CSAM” by these dbs, then you do the same thing with that image before it gets deleted.
To avoid false positives, you also train the AI on general non-CSAM content.
Ideally this process would be fully automated so no-one has to look at that shit - over time, ypu’d theoretically get a neural net capable of identifying CSAM reliably with few or no false positives or false negatives .. Admins could also try for some kind of distributed training, where each contributes weight deltas from local training, or each builds up LoRA-style improvement modules and people combine them to reduce bandwidth for modification sharing.
Is there a reason it needs a PK vs just being able to point it at a local folder and running as a user with write access?
Most lemmy servers do not have a GPU, so that would make it too slow. But I plan to provide this option soon if people want it.
I think there’s potential here for a centralised database of metadata (hashes, time-stamps, usernames) so it’s easier to stamp this out as a connected community. As has been noted, as long as you’re working in good faith, most government authorities won’t prosecute people running platforms like a Lemmy instance. But joining together to show solidarity might make it easier to prove that good faith, especially for smaller instances.
The tools has a lot of false positives as it casts a wide net to catch the actual positives. There’s no human review involved, so you can’t use these hits to mark usernames and time-stamps unless someone goes thjrough the hits and finds actual CSAM.
That’s reasonable, I wouldn’t want anyone to have to do those manual checks. However, wouldn’t false positives be relatively rare for a single username? Could correlational data of repeated flags within a certain timeframe be used to single out bad actors?
Yes, correlating single usernames would be a good way, but currently the script cannot connect images to posts
