Personal photos of Australian children are being illicitly used to train artificial intelligence (AI) tools, Human Rights Watch (HRW) reported on Tuesday. According to the report, these images are harvested from the internet without the children’s or their families’ consent, then incorporated into data sets that various companies use to train their AI models. HRW previously reported similar concerns and emphasized the potential misuse of these AI tools to generate harmful deepfakes.
HRW identified identifiable photos of Australian children in LAION-5B, a data set used to train major AI models. These pictures sometimes include personal details such as names, ages, and locations, making the children easily identifiable. The HRW analysis revealed 190 photos of children spanning all Australian states and territories, emphasizing that many of these images held a measure of privacy as they weren’t accessible via online searches or public websites.
LAION, the German non-profit managing the data set, validated HRW’s findings and committed to removing the personal photos from their data set as of June 1. HRW expressed concerns over the potential for AI models to perpetuate privacy risks as they can replicate the sensitive data they were trained on. The organization’s report highlights past incidents where AI models inadvertently leaked personal data.
Additionally, HRW warned that malicious actors have exploited LAION-trained AI tools to produce explicit imagery of children, significantly increasing the risks associated with deepfakes. This amplified threat has led to Australia’s introduction of the Criminal Code Amendment (Deepfake Sexual Material) Bill to criminalize the creation and sharing of non-consensual sexually explicit deepfake content. However, this legislation primarily pertains to adults, with existing child abuse material laws covering similar content involving minors.
In anticipation of the Australian government’s upcoming Privacy Act reforms in August, which include a Children’s Online Privacy Code, HRW urges the government to enact comprehensive laws to safeguard children’s data from AI misuse. HRW’s recommendations include prohibiting the scraping of children’s data for AI applications and non-consensual digital replication of children’s likenesses. The organization also advocates for processes to attain justice for affected children.
For more details, you can read the full article on JURIST.