Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
-
@troed @GossiTheDog plot twist of the year would be if the "dataset" they're talking about turned out to be "any image file uploaded to an S3 bucket between 2022 and today"
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog I didn't have "CSAM at scale is unavoidable" on my 2026 bingo card.
-
@masek @GossiTheDog But have they plundered Amazon S3 customer data, that the customers had set as private ?
@imbrium_photography @masek @GossiTheDog - I like the word that you have used: "Plundered" Private Data that was set to privacy.
-
@GossiTheDog @scottgal they say they're not training on it, it was detected before training. But that's not the point. Amazon got the stuff from somewhere, and a decent person would report where it came from so that the rozzers can trace it back upstream. I flat out don't believe Amazon's claim to not know where it came from, they must know, because they must have got copyright clearance for making a derivative work from all that content
@DrHyde @GossiTheDog @scottgal - Or Plundered Data.
-
@GossiTheDog BUT certain types of AI it would be obviously. THOSE need to exist in a regulated way and made open source. Like current PII scrubbing models it's a public good but I don't know any commercial company who COULD do it. Orthogonal sorry but just occurred to me...how do you get those models?
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-dataIf you’re using generative
AI toolsapplied statistics, there’s a pretty good chance you’regenerating imagery withsupporting the distribution of child porntraining data behind the scenes. -
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-dataALT TEXT:
Bloomberg
Amazon Found 'High Volume' Of Child Sex Abuse Material in AI Training Data.
The tech giant reported hundreds of thousands of cases of Child Sex Abuse Material but won’t say where it came from. -
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog That article is full of red flags from Amazon. They claim they have a "lower threshold" so they're "overreporting" but not providing info on the source of the images?
That sounds like they're trying to break NCMEC's reporting system either through malice or incompetence.
Also it sounds like they're not keeping the provenance of the data they're using - which strongly suggests that they're not obtaining that data in a legal manner
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog and, every one of those pictures has been seen and classified by a minimum-wage worker in the third world so that the user doesn't get to see it (at a predictable cost to said third-world worker's mental health).
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog 'is refusing to tell regulators'?
Good luck with that if there are any datasets in the UK. Time for arrests and seizure of machines.
It should be the same in the US, but of course nothing comes before the 'mighty' dollar
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-dataThe sets are what they stole from billionaires and senators' sons. Even themselves.
What the fuck is wrong with people? No one gets to convince me we ain't the worst disease this planet must suffer.
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog I wonder if they found the data crawling their user storage and they don't want to tell about it to keep the patient money money
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog "refusing to tell regulators which data sets"
In what world is this not criminal, and why are we living in that one?
-
well there's your Epstein files right there!
@Sassinake @GossiTheDog Scraped from a DoJ server left unsecured by DOGE? Everything's possible with these people
-
Amazon have reported "hundreds of thousands" of pictures of child sexual abuse material found in shared AI training data... but is refusing to tell regulators which data sets.
If you're using generative AI tools, there's a pretty good chance you're generating imagery with child porn training data behind the scenes.
https://www.bloomberg.com/news/features/2026-01-29/amazon-found-child-sex-abuse-in-ai-training-data@GossiTheDog Famously, generative AI has been hilariously bad at producing a picture of a glass of wine that's anything other than about half full. Ask for one that's full or nearly empty and it can only show you ones that match it's training data: where all the glasses show a tasteful measure. And good luck asking for a clock face that doesn't show seven minutes past ten. It just can't extrapolate. However ask it what a naked child look like and it's remarkably good at it. Why? Well ask the people who tripped CSAM filters by downloading image training data. Dear Elon, why is Grok so good at making child porn. Did you train it on your own kids or ours? And telling the interface not to show you the filthy kiddie pics that it's gathered, is a bit like selling a porn magazine and asking customers not to look at pages 12-27 because you accidentally abused some kids when you made it.
-
P pelle@veganism.social shared this topic
