Paedophiles are exploiting artificial intelligence (AI) technology to generate and sell life-like child sexual assault content. Some people gain access to the photographs by subscribing to accounts on popular content-sharing websites like Patreon.
What really happened?
- The BBC has discovered that paedophiles are exploiting artificial intelligence (AI) technology to produce and market lifelike child sexual assault content. Some people are getting access to the photographs via subscribing to accounts on popular content-sharing websites like Patreon. According to Patreon, its website has “zero tolerance” for such imagery.
- The makers of the abused photos use artificial intelligence (AI) software called Stable Diffusion, created to produce a graphic design or art images. Thanks to AI, computers may now carry out activities that would normally require human intelligence.
- The Stable Diffusion software enables users to describe whatever image they desire using word prompts, and the program automatically generates the desired image.
Three stages sharing of images
- Paedophiles use AI software to create their photos.
- They advertise images on websites like the Japanese photo-sharing service Pixiv.
- These accounts include links that take users to their more graphic content, which can be viewed for a fee on accounts on websites like Patreon.
Some of the picture makers share their work on Pixiv, a well-liked Japanese social networking site that producers of manga and animation primarily use. The creators promote it to market their work in groups and using hashtags, which index subjects using keywords. Still, because it is in Japan, where it is permitted to share sexualised cartoons and drawings of minors, they are able to do so.
The artificial intelligence revolution has resulted in a flood of shockingly lifelike visuals depicting child sexual exploitation, raising fears among child-safety investigators that they may undermine attempts to discover victims and stop real-world abuse.
Thousands of AI-generated child-sex photographs have been discovered on forums spanning the dark web, a layer of the internet accessible only through special browsers, with some users offering extensive instructions on how other paedophiles might create their own masterpieces. Children’s images, including the content of known victims, are being repurposed for this evil output.
New AI- Tool
The new AI tools, known as diffusion models, enable anyone to build a convincing image by simply entering a brief description of what they wish to see. The models, which included DALL-E and Stable Diffusion, were fed billions of photographs collected from the internet, many of which included real children and were obtained from photo sites and personal blogs. They then replicate those visual patterns in order to build their own visuals.
The tools have been praised for their visual ingenuity. They have been used to win fine arts contests, illustrate children’s books, produce false news-style images, and make synthetic pornography of non-existent figures dressed as grownups.
Safeguarding tool against CSAM
- A new AI-powered tool promises to be 99% accurate in detecting child abuse photographs.
- The non-profit Thorn created the Safer tool to help firms who do not have in-house filtering systems find and remove such images.
Safer’s detection services include
- Image Hash Matching is the company’s flagship service, which generates cryptographic and perceptual hashes for images and compares them to known CSAM hashes. At the time of publication, the database had 5.9 million hashes. To protect user privacy, hashing takes place in the client’s infrastructure.
- CSAM Image Classifier is a machine learning classification model that is used within Safer to forecast whether a file is CSAM. The classifier was trained on datasets including hundreds of thousands of photos, including adult pornography, CSAM, and other imagery, and it can help identify potential new and unknown CSAM.
- Video hash matching: Service that generates cryptographic and perceptual hashes for video scenes and compares them to hashes representing suspected CSAM scenes. The database currently contains over 650k hashes of probable CSAM scenes.
- SaferList for Detection: A service that allows Safer customers to harness the collective knowledge of the Safer community by matching hash sets given by other Safer customers to widen detection attempts. Customers can choose which hash sets they want to include.
The CSAM is increasing very rapidly, and with the use of AI tools, it becomes effortless for bad actors to create child abuse images. Regarding child safety specialists, Users have publicly discussed tactics for creating obscene photographs and avoiding anti-porn filters on the dark-web paedophile forums, including utilising non-English languages they believe are less prone to suppression or discovery. What is most dangerous is the open release of these tools on the web. There is no straightforward, coordinated strategy to remove such decentralised rogue actors.
Authors: Himanshi Singh Associate, Policy & Advocacy team