Personal pictures of Brazilian children have been misused to power artificial intelligence tools, according to a report released by Human Rights Watch (HRW) on Monday. These images are being included without the children’s or their parents’ knowledge or consent in a dataset used to train AI models. The AI tools derived from this data can create deepfake images using the children’s likenesses, which HRW points out may expose these children to significant risks of exploitation and harm.
The analysis conducted by the NGO reveals that a popular dataset used for AI training, known as LAION-5B, contains links to photos of Brazilian children. Some of these photos come with detailed captions and URLs, which may include personal data such as the child’s name, the hospital where they were born, and their location at the time the photo was taken. HRW found 170 such photos from ten different states in Brazil, noting that they only reviewed a minuscule fraction—0.0001 percent—of the total dataset.
HRW Brazil emphasized on X that neither the children in these images nor their parents were aware of their likeness being used for AI purposes. Hye Jung Han, a HRW researcher specializing in children’s rights and technology, discussed the findings in an interview with the Brazilian newspaper Folha de S.Paulo, stating that the government should urgently adopt policies to protect children’s data from AI misuse.
The dataset appears to contain private images, many of which were initially posted on personal blogs that are not easily accessible through simple online searches. These AI tools are capable of generating hyper-realistic images by learning from these likenesses, which can then be misused to create explicit imagery of the children involved.
In response to these findings, Brazilian lawmakers have proposed regulations to ban the use of AI technologies that generate sexually explicit images of children without consent. Han asserts that incorporating rigorous data privacy protections in AI regulations will be crucial for safeguarding children’s rights as AI technology continues to develop.
LAION, the nonprofit managing the dataset, acknowledged the allegations and announced their intention to remove the images of children. They further advised that the best prevention method against such misuse is for guardians to avoid posting personal photos of children online.
Further details can be found on JURIST.