cross-posted from: https://lemmy.today/post/25826615
For those not familiar, there are numerous messages containing images being repeatedly spammed to many Threadiverse users talking about a Polish girl named “Nicole”. This has been ongoing for some time now.
Lemmy permits external inline image references to be embedded in messages. This means that if a unique image URL or set of image URLs are sent to each user, it’s possible to log the IP addresses that fetch these images; by analyzing the log, one can determine the IP address that a user has.
In some earlier discussion, someone had claimed that local lemmy instances cache these on their local pict-rs instance and rewrite messages to reference the local image.
It does appear that there is a closed issue on the lemmy issue tracker referencing such a deanonymization attack:
https://github.com/LemmyNet/lemmy/issues/1036
I had not looked into these earlier, but it looks like such rewriting and caching intending to avoid this attack is not occurring, at least on my home instance. I hadn’t looked until the most-recent message, but the image embedded here is indeed remote:
https://lemmy.doesnotexist.club/pictrs/image/323899d9-79dd-4670-8cf9-f6d008c37e79.png
I haven’t stored and looked through a list of these, but as I recall, the user sending them is bouncing around different instances. They certainly are not using the same hostname for their lemmy instance as the pict-rs instance; this message was sent from nicole92 on lemmy.latinlok.com, though the image is hosted on lemmy.doesnotexist.club. I don’t know whether they are moving around where the pict-rs instance is located from message to message. If not, it might be possible to block the pict-rs instance in your browser. That will only be a temporary fix, since I see no reason that they couldn’t also be moving the hostname on the pict-rs instance.
Another mitigation would be to route one’s client software or browser through a VPN.
I don’t know if there are admins working on addressing the issue; I’d assume so, but I wanted to at least mention that there might be privacy implications to other users.
In any event, regardless of whether the “Nicole” spammer is aiming to deanonymize users, as things stand, it does appear that someone could do so.
My own take is that the best fix here on the lemmy-and-other-Threadiverse-software-side would be to disable inline images in messages. Someone who wants to reference an image can always link to an external image in a messages, and permit a user to click through. But if remote inline image references can be used, there’s no great way to prevent a user’s IP address from being exposed.
If anyone has other suggestions to mitigate this (maybe a Greasemonkey snippet to require a click to load inline images as a patch for the lemmy Web UI?), I’m all ears.
Could it potentially be enough to find location - like even if not city, then state or at least country?
And ofc not just these Nicole pics, but any pics at all, across the entire Fediverse. Worse, upload it via posting to a small community with like 5-10 subscribers and get the IPs of all of those who see the content (by downloading the image from your self-hosted server), then correlate with comments in it to map to usernames (I mean narrow down the list to those 5-10 accounts).
I suppose it is fortunate that there aren’t any totalitarian regimes anywhere in the world that might be interested in keeping tabs on who isn’t using corporate enshittified platforms… Like surely Musk won’t deny visas to people in the USA who use Lemmy, r-r-right??? (Or deny employment even to people working for corporations that even so much as have a contract with the USA government, regardless of whether the person in question is actually working on it or not, or are even aware that their company has such contracts at all?).
I think we may need to expect the worst, moving forward, then be pleasantly surprised if it doesn’t happen, rather than 100% count on the best happen for certain, like our very lives depended upon it.
@rimu@piefed.social how does PieFed fare in this regard?
Location would be possible. For me it’s a few 100km off, but usually the GeoIP databases are more accurate.
Piefed doesn’t do much image caching or proxying. It only keeps thumbnails around. Once you open a post with more than a thumbnail in it (a full picture), your IP is revealed to the image hoster.
Not great. PieFed does not make a local copy of inline images, like Lemmy sometimes does.
This makes me worried to click on certain post titles that I’ve seen lately… Edit: on the bright side, PieFed doesn’t include words in its titles, just numbers, so that might offer a bit of protection, for keyword scanners built to work for Lemmy.
Would using a proxy be sufficient?