Australian children’s photos have been included in an AI image-generating tool dataset without their knowledge or consent. An analysis of 0.0001% of the 5.85 billion images contained in the Laion-5B dataset found 190 photos of Australian children scraped from the internet. The photos were easily identifiable, with some names and information about when and where they were taken included.
The children’s images were lifted from photo and video sharing sites, as well as school websites. One photo featured two boys in front of a colourful mural, revealing their names, ages, and preschool attended. The dataset also included an unlisted YouTube video of schoolies celebrations.
Laion, the organisation behind the dataset, stated that any material related to children’s images came from links pointing to publicly available sections of the internet. However, HRW found that this practice risks harming two groups of children: those whose photos are scraped and those who potentially have malicious AI tools built on the dataset used against them.
HRW is calling for legislation to update the Privacy Act to prohibit scraping of children’s data into AI and prohibit nonconsensual digital replication or manipulation of their likeness.