Artificial intelligence-generated images of child sex abuse are becoming "your worst nightmare" and threaten to overwhelm the internet, a security watchdog has warned. The Internet Watch Foundation (IWF) said it had discovered nearly 3,000 AI-generated abuse images that clearly broke UK law.
The UK-based organization said existing real-life images of abuse victims are being fed into AI models, which then generate new depictions.
The group added that the technology was also used to create images of celebrities who were "de-aged" and then depicted as children in scenes of sexual abuse. Other examples of child sexual abuse material (CSAM) include using artificial intelligence tools to "nude" images of clothed children found online.
The International Women's Forum warned this summer that evidence of abuse caused by AI was starting to emerge, but it said the latest reports showed the use of the technology was accelerating. IWF chief executive Susie Hargreaves said the watchdog's "worst nightmare had come true".
"Earlier this year, we warned that AI images could soon become indistinguishable from real photos of children being sexually abused, and that we might start to see such images proliferating in greater numbers," she said. "We are now past that stage. Chillingly, we are seeing criminals deliberately train their AI on images of real victims who have been abused. Children who have been raped in the past are now being included in new scenarios because someone, somewhere, wants to see it."
The International Women's Forum said it had also seen evidence of AI-generated images being sold online. The latest findings are based on a month-long investigation into dark web child abuse forums. It investigated 11,108 images on the forum, 2,978 of which broke UK law by depicting child sexual abuse.
AI-generated CSAM is illegal under the Child Protection Act 1978, which criminalizes the taking, distribution and possession of "indecent or pseudo-photographs" of children. The International Women's Forum said the vast majority of illegal material it found breached the Child Protection Act, with more than a fifth of the images classified as Category A, the most serious category of content that can depict rape and sexual abuse.
The Coroners and Justice Act 2009 also criminalizes non-photographic prohibited images of children, such as caricatures or drawings. The IWF fears that a wave of AI-generated CSAMs will distract law enforcement agencies from detecting real abuse and helping victims.
"This material has the potential to overwhelm the Internet if we don't control this threat," Hargreaves said.
Dan Sexton, IWF's chief technology officer, said the image generation tool StableDiffusion, a publicly available AI model that can be tweaked to help generate CSAM, was the only AI product discussed on the forum.
"We are seeing discussion around creating content using StableDiffusion, which is publicly available software."
StabilityAI, the British company behind StableDiffusion, said it "prohibits misuse on our platform for any illegal or unethical purpose, and our policies clearly state that this includes CSAM."
The government says a cybersecurity bill set to become law will cover AI-generated CSAM and require social media companies to prevent it from appearing on their platforms.