Paedophiles are using AI programs to generate lifelike images of child sexual abuse, prompting concerns among child-safety investigators that they will undermine efforts to find victims and combat real-world abuse.
According to a report by the Washington Post, the rise of AI technology has sparked a ‘predatory arms race’ on paedophile forums across the dark web.
The makers of the abusive images use software called Stable Diffusion, which was intended to generate images for use in art or graphic design.
But predators have been using the software to create their own realistic images of children performing sex acts, and have been sharing detailed instructions for how other paedophiles can create their own.
‘Children’s images, including the content of known victims, are being repurposed for this really evil output,’ said Rebecca Portnoff, director of data science at nonprofit child-safety group Thorn.
‘Victim identification is already a needle-in-a-haystack problem, where law enforcement is trying to find a child in harm’s way,’ she told the Post. ‘The ease of using these tools is a significant shift, as well as the realism. It just makes everything more of a challenge.’
In the UK, a ‘pseudo image’ generated by a computer which depicts child sexual abuse is treated the same as a real image and is illegal to possess, publish or transfer.
The National Police Chiefs’ Council (NPCC) lead on child safeguarding, Ian Critchley, said it would be wrong to argue that because no real children were depicted in such ‘synthetic’ images.
He told the BBC that the programs could allow paedophiles to ‘move along that scale of offending from thought, to synthetic, to actually the abuse of a live child’.
The emergence of such images also has the potential to undermine efforts to find victims and combat real abuse, forcing law enforcement to go to extra lengths to investigate whether a photograph is real or fake.
According to the publication, AI-generated child sex images could ‘confound’ the central tracking system built to block such material from the web because it is designed only to catch known images of abuse, rather than detect newly-generated ones.
Law enforcement officials, who work to identify victimised children, may now be forced to spend time determining whether the images are real or AI-generated.
AI tools can also re-victimize any individual whose photographs of past child sex abuse are used to train models to generate fake images.
Some of the image creators are posting on a popular Japanese social media platform called Pixiv, which is mainly used by artists sharing manga and anime.
But because the site is hosted in Japan, where sharing sexualised cartoons and drawings of children is not illegal, creators can share their work via groups and hashtags.
The subscription-based platform Patreon is also used to host the obscene images, with accounts selling AI-generated, photo-realistic images of children behind a paywall with different levels of pricing depending on the type of material requested.
Journalist Octavia Sheepshanks told the BBC her research suggested users appeared to be making child abuse images on an industrial scale.
‘The volume is just huge, so people [creators] will say “we aim to do at least 1,000 images a month,”’ she said.
‘Within those groups, which will have 100 members, people will be sharing, “Oh here’s a link to real stuff,” she added.
‘The most awful stuff, I didn’t even know words [the descriptions] like that existed.’
On dark-web paedophile forums, users openly discussed strategies for how to create explicit photos and dodge anti-porn filters, which include using non-English languages they believe are less vulnerable to suppression or detection.
According to WaPo, one forum with 3,000 members saw roughly 80 percent of respondents to a recent internal poll say they had used or intended to use AI tools to create child sexual abuse images.
Forum members also discussed ways to create AI-generated selfies and build a fake school-age persona in hopes of winning children’s trust, the publication reported.
Ms Portnoff said her group also has seen cases in which real photos of abused children were used to train the AI tool to create new images showing those children in sexual positions.
A spokesperson for Patreon said it had a ‘zero-tolerance’ policy towards hosting child abuse images, real or otherwise.
‘We already ban AI-generated synthetic child exploitation material,’ it said, describing itself as ‘very proactive’, with dedicated teams, technology and partnerships to ‘keep teens safe’.
A spokesman for Pixiv also said it placed immense emphasis on addressing this issue, and had banned all photo-realistic depictions of sexual content involving minors on May 31.