How do you find an illegal image without looking at it? (mahmoud-salem.net)

This matters greatly if you want to self-host something like Matrix, and you permit federation..

You WILL get a CSAM spam issue. It will get caught in your server cache. And you won't catch it until after the fact. And shit admin tools will not properly remove the spammer or content.

Better yet, if you run Matrix, disable image caching and preloading.

What does it say about us, as a society, or just as _humans_, where the scale and magnitude of this problem is so great and only growing? Where and how are we failing ourselves that the sort of mental illness that percolates and drives this sort of behavior festers, amplifies, and converts into actual, illicit action?

These numbers are mind-boggling, and while I understand that a "few (extremely) bad apples" are probably responsible for an outsized amount of production, AND that AI-generated imagery is flooding the zone disproportionate to the amount of actual human children being physically harmed, it's still absolutely wild to me that we collectively are producing and consuming so much of this content, despite it being largely universally considered essentially the most abhorrent thing possible.

What would fixing this at the root cause even start to begin? How do we apply whatever combination of therapeutic intervention or further societal pressure or whatever might work to reduce the incidence of people having these urges, exploring them, feeding them, and sometimes acting on them? We see signs in every airport bathroom telling us to look for signs of trafficking. Trafficking intervention training is a huge deal in the travel industry in general. There are early intervention and detection systems for social workers and case workers.

But has anyone spent any real time looking at this from the other side: the side of the offender? I imagine there's research on the typical chain of how someone gets "onboarded" here: it probably starts with some early abuse, or if not that, early exposure or early curiosity, and then snowballs from there. I'm just thinking out loud about how large the magnitude of the problem is on the offender side if we're talking about this volume of images, and how we might be able to evaluate things from the "ounce of prevention worth a pound of cure" side of things, because damn is this depressing.

[dead]

[dead]

> no X. no Y. just Z

i am so sick of AI slop writing..

[dead]

“ Over 1.5 million of those reports involved generative AI. Some of this material depicts entirely fictional children. But a growing share is generated using the likenesses of real, identifiable children — children who have never suffered contact abuse, but who are now victims nonetheless. And all of it — real or synthetic — floods into the same investigation pipeline, where human analysts must treat every image as potentially depicting a real child in danger.”

If any of the leading AI companies are looking to get back in the good graces of the public, they should seriously think about releasing an open source model that reliably labels media (text, photo or video) with a probability said media is AI generated.

There is a 0% chance they don’t already have models for this to prevent feeding their models AI generated training data. So release it.

simple, capture people who are already seeking these images, and keep them somewhere in confinement, but with access to the internet, they find more and act as agents for society for life, be good little perverted monsters, and they dont get castrated and released into the general prison population.

Haven't read the post yet but I think the general technique is variations on spectral analysis. Break up the image into spectral components & then figure out a relative similarity metric based on spectral statistics.

Edit: That's exactly what they do. Basic stuff if you know the fundamental techniques.

This is one of the most legible, well-detailed, and well-written article I've seen on perceptual hashing. It must have taken months of effort to pull off, and I'd love to see the author write about other things.

But the article fails to take its statements to their logical conclusion, in one section, he writes,

    > Every false positive means an innocent person's content was flagged — a family photo, a medical image, a piece of art. It means unnecessary investigation, potential harm to reputation, and erosion of trust in the system. At scale, even a 0.01% false positive rate means thousands of wrongful flags per day.

and,

    > In practice, the industry errs heavily toward minimizing false negatives — catching every possible match — and then uses human review to resolve false positives. This means the system flags aggressively but confirms carefully. The cost of a false positive is an investigation. The cost of a false negative is a child.
    > 
    > This is also why the hybrid approach from Chapter VI matters. Perceptual hashing against a verified database has a low false positive rate — but not zero. Certain images (blank, solid-color, simple gradients) produce hashes that collide with database entries by coincidence, not because they depict abuse. Production systems include collision detection to filter these out before matching. Classifiers for unknown material have a higher false positive rate still (the model is making a judgment, not a comparison). By layering them — hashing first, then classifiers, then human review — the system can be both aggressive and precise. But no layer is perfect, and the threshold remains a human decision.

If there is a way to "include collision detection to filter these out before matching" then why do they "then human review?" The author starts the next section with, "Three Steps. No One Sees the Image."

But they do human review to eliminate false positives? Both statements can't be simultaneously true - "no human ever sees it," or "by layering them — hashing first, then classifiers, then human review — the system can be both aggressive and precise."

Secondly, although I'm not a researcher, I think I and a lot of researchers would love to see this "aggressive, but precise algorithm" that eliminates collisions (an imprecise term - while here it means an image of a background or a setting that ticks off the similarity system; it's still not exactly a collision in the classical sense as the algorithm is a type of clustering with hashes) without making the algorithm useless? As far as I'm aware, no such algorithm exists without either becoming useless or having significant false positives. But I might be wrong.

At one point in the article, the author says, "The cost of a false negative is a child." This "aggressive and precise" system diverts resources from actual investigations and prosecution. A few examples,

A very famous case from 2022, https://www.nytimes.com/2022/08/21/technology/google-surveil...

A more precise example, as the author mentions PhotoDNA,

    > LinkedIn found 75 accounts that were reported to EU authorities in the second half of 2021, due to files that it matched with known CSAM. But upon manual review, only 31 of those cases involved confirmed CSAM. (LinkedIn uses PhotoDNA, the software product specifically recommended by the U.S. sponsors of the EARN IT Bill.) 

PhotoDNA's "aggressive and precise" have a 58.6% false positive rate when tested. That means nearly 60% of the cases it generates for investigations wasted investigators time, leading to fewer investigations overall.

from,