[ad_1]
(Sydney) – Personal photos Australia Children are being used to create powerful artificial BroadCast Unitedligence (AI) tools without their knowledge or consent or that of their families, Human Rights Watch said today. The images are scraped from the internet and deposited into a massive dataset that companies then use to train their AI tools. In turn, others use these tools to create malicious deepfakes, putting more children at risk of exploitation and harm.
“No child should have to live in fear that their photos could be stolen and used as a weapon,” Han Hye-jungchildren’s rights and technology researcher and advocate at Human Rights Watch. “The Australian government should urgently pass laws to protect children’s data from being misused by artificial BroadCast Unitedligence.”
The Human Rights Watch analysis found that the LAION-5B dataset, which is used to train popular artificial BroadCast Unitedligence tools and was built by scraping large portions of the internet, contained links to identifiable photos of Australian children. Some of the children were named in the accompanying captions or in the URLs where the images were stored. In many cases, their identities were easily traceable, including information about the time and location of the children when the photos were taken.
In one photo, two boys, aged 3 and 4, grin from ear to ear as they hold paintbrushes in front of a colorful mural. The caption lists the children’s full names and ages, as well as the name of the kindergarten they attend in Perth, Western Australia. There appears to be no information about the children available on the internet.
Human Rights Watch found 190 images of children from all states and territories of Australia. This is likely a significant underestimate of the amount of personal data on children in LAION-5B, as Human Rights Watch reviewed less than 0.0001% of the 5.85 billion images and captions contained in the dataset.
The photos reviewed by Human Rights Watch cover the entirety of childhood. They capture the intimate moment of a baby being born into a doctor’s gloved hands and connected to its mother by an umbilical cord; toddlers blowing bubbles or playing musical instruments in kindergarten; children dressing up as their favorite characters for Book Week; and girls in bathing suits at a school swimming carnival.
The photos also include Aboriginal children, including those from the Anangu, Arrernte, Pitjantjajara, Pintupi, Tiwi and Warlpiri tribes mentioned in the title. There are toddlers dancing to songs in Aboriginal languages; a girl proudly holding a sand lizard by the tail; and three young boys, wearing traditional body paint, holding hands.
Many of these photos were originally seen by only a few people and were previously private. They do not appear to be found by online searches. Some of the photos were posted by children or their families on personal blogs and photo and video sharing sites. Others were uploaded by schools, or by photographers hired by the families to capture personal moments and portraits. Some of these photos are not available on the public versions of these sites. Some were uploaded years or even decades before the LAION-5B was created.
Human Rights Watch found that LAION-5B also contains some photos of children whose privacy has been protected. One photo, a close-up of two boys making faces, was taken from a YouTube video of teenagers celebrating graduation week after final exams. The creator of the video took precautions to protect the privacy of the people in the video: The privacy setting is set to “unlisted,” and the video does not appear in YouTube search results.
YouTube Terms of Service Scraping or collecting potentially personally identifiable information, including images of their faces, is prohibited except in certain circumstances; this situation appears to violate these policies. YouTube did not respond to our request for comment.
Once their data is collected and fed into AI systems, these children’s privacy is further threatened by technical flaws. AI models, including those trained on LAION-5B, are notorious for leaking private information; they can Make an identical copy The materials they were trained on include Medical Records Guardrails that some companies put in place to prevent sensitive data from being leaked have been repeatedly shattered.
Furthermore, current AI models are unable to forget the data they were trained on, even if that data is later removed from the training dataset. This risk of permanence would be particularly harmful to Indigenous Australians, many of whom restrict the reproduction of images of deceased people during periods of mourning.
Human Rights Watch said these privacy risks pave the way for further harm. By training on photos of real children, AI models are able to few Photos, even Single imageMalicious actors use LAION-trained AI tools to Produces a clear image Harmless photos of children are used, as well as explicit images of child survivors who have been sexually abused Scratched LAION-5B.
Similarly, the presence of Australian children in LAION-5B contributes to the likelihood that AI models trained on this dataset will generate realistic images of Australian children. This significantly increases the existing risk that children face that someone will steal their likeness from photos or videos they post online and use AI to manipulate them into saying or doing things they have never said or done.
In June 2024, about 50 Melbourne girls reported that photos on their social media profiles were taken and processed by artificial BroadCast Unitedligence to create pornographic deepfake videos of them, which were then circulated online.
Fictional media has always existed, but producing it requires time, resources, and expertise, and is largely impractical. Current AI tools can create lifelike output in seconds, are often free, and are easy to use, threatening to create a flood of non-consensual deepfakes that could circulate online forever, causing lasting harm.
LAION, the German nonprofit that manages LAION-5B, confirmed on June 1 that the dataset contained personal photos of children discovered by Human Rights Watch and pledged to remove them. The organization disputed the claim that AI models trained on LAION-5B could reproduce personal data verbatim. “We urge Human Rights Watch to contact these individuals or their guardians to encourage them to remove this content from the public domain, which will help prevent its recirculation,” LAION said.
Australian Attorney General Mark Dreyfus recently Introduction A bill In Parliament, a ban on the creation or sharing of adult pornographic deepfakes without consent was introduced, noting that such images of children would continue to be considered child abuse material under the Criminal Code. However, Human Rights Watch said this approach ignores the deeper problem that children’s personal data remains unprotected from misuse, including the manipulation of real children’s likenesses into any form of deepfake video without consent.
In August, the Australian government will reform the Privacy Act, including drafting Australia’s first child data protection law, the Children’s Online Privacy Act. Human Rights Watch said the law should protect the best interests of children as recognized in the United Nations Convention on the Rights of the Child and all rights that children have in relation to the collection, processing, use and retention of their personal data.
The Children’s Online Privacy Act should prohibit the scraping of children’s personal data into artificial BroadCast Unitedligence systems. It should also prohibit the digital copying or manipulation of children’s likenesses without consent. And it should provide mechanisms for children who have been harmed to seek meaningful justice and redress.
The Australian government should also ensure that any proposed AI regulations include data privacy protections for everyone, especially children.
“Generative AI is still an emerging technology, and the associated harms that children have already experienced are not inevitable,” Han said. “Protecting children’s data privacy now will help shape the development of this technology into one that promotes children’s rights rather than violates them.”
[ad_2]
Source link
