Photographs of Australian children have been used to train generative AI tools without their knowledge or consent, this is according to a new report from the Human Rights Watch (HRW).
Human Rights Watch claims these images of minors have been scraped by AI companies and used by companies to train their AI tools.
The HRW found these photographs are being accessed online and added to the popular open-source LAION-5B dataset. This dataset, which contains 5.85 billion images, includes URL links to their source on the internet and reportedly contains easily traceable information, such as when and where the photo was taken.
One of the photos the organisation analysed included two boys, aged between three and four, holding paintbrushes before a mural. The name of their preschool, located in Perth, was traceable via LAION-5B, along with the full names of these children.
The HRW claims that they found 190 photos of children from all over Australia and it says this is likely to be a significant undercount. Children of all ages and from First Nations communities were counted in the HRW’s review of the AI model which included such photos as kids at swimming carnivals, and kids dressed for Book Week.
Many of the photos the HRW spotted were originally posted in capacities where their visibility was measured by privacy, and are unable to be found via a standard online search. Some photos were uploaded by schools or family-hired photographers. One photo observed was traceable to an ‘unlisted’ video on YouTube, which is a violation of YouTube’s policies on data scraping.
“Children should not have to live in fear that their photos might be stolen and weaponised against them,” children’s rights and technology researcher and advocate at Human Rights Watch Hye Jung Han said. “The Australian government should urgently adopt laws to protect children’s data from AI-fueled misuse.”
The HRW warns there are extreme privacy concerns with this kind of conduct from all generative AI companies. Bad actors could use such tools to create malicious or explicit imagery.
LAION told the HRW that it would remove the photos of the children identified. The AI company also claimed that children and their guardians were responsible for removing such photos from the internet, with the company arguing this to be the best protection against misuse.
But the damage could already be done, with the HRW claiming that AI models can’t forget the data that they’re being trained on.
Next month, the Australian Government is set to introduce updates to the Privacy Act, including a new protection for children, called the Children’s Online Privacy Code. The HRW argues that this code should include protections for children against AI, to prohibit companies from accessing photos of kids and using them for AI systems.
“Protecting children’s data privacy now will help to shape the development of this technology into one that promotes, rather than violates, children’s rights,” Han added.
It’s an extremely worrying report that the Human Rights Watch has put out, but unfortunately, it’s far from the first, and it’s a continuing trend for AI companies to keep bending the rules to their will. Last Month, Perplexity AI found itself in hot water, after an investigation revealed that it was subverting the Robots.TXT clause, and was scraping web content regardless. Privacy violations and non-answers on where content is sourced from are common among AI companies.
Gizmodo Australia has reached out to LAION for comment.
If you or someone you know has been the victim of a crime, please call 000. If you require support, know that help is available. Please call Lifeline at 13 11 14, or Kids Helpline at 1800 55 1800.
Image: iStock