AI Models Are Using Protected Photos of Children for Training

A recent report has uncovered that hundreds of photos of children are present in widely used databases for training AI models and image generators.

Despite strict privacy laws protecting your child's photos, a new report suggests that AI models are likely still using them for training.

The constant stream of negative press about AI shows no signs of stopping. It seems like every other day brings another story about an AI mistake leading to real-world problems for regular people.

Recently, it's come to light that AI models are indeed gathering all the information available online, including photos of children from various sources on the internet.

Human Rights Watch Reveals AI is Using Children's Photos for Training

A report from Human Rights Watch (HRW) discovered 190 photos of Australian children in a widely used database of online screenshots that train AI models, including popular image generators.

This is the second report on AI training using children's photos. In June, HRW released a similar report highlighting 170 photos of Brazilian children being used online.

"Children should not have to live in fear that their photos might be stolen and weaponized against them. The Australian government should urgently adopt laws to protect children’s data from AI-fueled misuse," said Hye Jung Han of Human Rights Watch.

Even more concerning, some photos included sensitive information such as the individual's name and location. Researchers found these photos were scraped from content with strict privacy settings, such as unlisted YouTube videos.

AI Training Is Forever

The database in question, LAION-5B, is managed by LAION, a non-profit, volunteer organization. LAION acknowledges that the misuse of children's photos to train AI models is a "larger and very concerning issue" and is committed to addressing these problems swiftly.

However, these efforts may not be enough to fully control the situation.

"Current AI models cannot forget data they were trained on, even if the data was later removed from the training data set," said Hye Jung Han of Human Rights Watch.

This means that even if a photo is completely erased from the internet and scrubbed from every database, it can still be accessed if an AI was trained on it.

The Legality of AI

Generative AI models are evolving at a rapid pace. Their swift integration into everyday technology has outstripped the ability of regulatory bodies to keep up.

However, legal actions are now beginning to address these issues. Australia is currently voting on reforms to the Children’s Online Privacy Code, which aims to include more protections against AI misuse. This is not an isolated case.

In the US, record labels like Sony, Universal, and Warner are suing AI music generators for copyright infringement, arguing that their models are trained on music owned by these labels.

The unregulated era of AI is coming to an end as people start to understand how these models operate. That is, assuming they don’t consume all our electricity first.