You’ve seen a headline saying your photos are being used to train AI, and now you’re wondering whether that includes your kids’ photos.
The short answer is yes, if the photos are public. But there are important caveats, and they give you more control than the headline suggests.
Here’s what Meta and Google do, what the law allows, and the few steps that protect your children.
The short answer
- Meta (Facebook, Instagram, Threads) uses public posts and photos to train its AI models, including Llama and Meta AI. Private posts and messages are treated differently.
- Google says it does not train its general AI models on your personal Google Photos library. Photos you post to the public web are a separate story.
- The bigger risk sits outside any single company: a public photo can be scraped by anyone, into datasets neither you nor Meta nor Google controls.
One word keeps repeating: public. That’s also the good news, because it points straight at the fix.
What Meta does with your photos
Meta has been open about the mechanism. It uses “information shared publicly” on its platforms to train generative AI (Meta). If your account is public, your posts, captions, and photos are in scope.
There’s a child-specific wrinkle. Meta says it doesn’t use data from accounts belonging to under-18s. But most young children don’t have their own account, they appear on their parents’ accounts. When pressed by regulators in Australia about whether photos of children on an adult’s public profile would be scraped, Meta confirmed they would (Malwarebytes). So the “we exclude minors” reassurance has a real gap: the photos of kids that parents post sit in the adult’s account, not the child’s.
Meta is one example of a wider pattern: the predictable result of a business model where public content is raw material. But it does mean the everyday act of posting your child to a public feed now has a second consequence you didn’t sign up for.
What about Google Photos?
Here the nuance matters.
Google has stated it does not train generative AI models outside of Google Photos on the personal content in your Google Photos library (Forbes). So the private backup of your camera roll sitting in Google Photos is not being mined to train Gemini. That’s a meaningful distinction, and fair to Google to note.
The catch is what happens outside that private library. A photo you upload to a public blog, a public Facebook post, an open forum, or a school’s public page lives on the open web, and the open web is exactly what large training datasets are built from. Where the photo was first stored matters far less than whether it ended up somewhere public.
The part worth worrying about: public datasets
“Which company can I trust?” is the wrong question, because a public photo escapes any single company’s control.
In 2024, Human Rights Watch found photos of identifiable children, including 170 photos of Brazilian children from at least 10 states and around 190 Australian children, scraped into LAION-5B, one of the large open datasets used to train image AI. The captions sometimes included the child’s name and location. Some of these photos had been posted years earlier, in low-key settings, by parents who never imagined them ending up in a machine-learning corpus (Human Rights Watch). Human Rights Watch also documented girls in Brazil targeted with sexually explicit deepfakes made with AI tools from photos taken from their social media.
None of this calls for deleting everything and panicking. It does mean treating a public photo as copyable, forever, by parties you’ll never identify. The same public exposure is what feeds facial-recognition systems, which we cover in sharenting and facial recognition. No opt-out form reaches all of them. Policing who trains on a public photo is a losing job; keeping the photo out of public view is not.
What the law lets them do (and where you have rights)
Your rights here depend a lot on where you live:
- EU and UK (GDPR). You have a right to object to your data being used for AI training. Meta says it honors objections submitted through its Privacy Center form.
- Brazil (LGPD). Brazilians have a comparable right to object, and Brazil’s data authority has been unusually assertive: in July 2024 the ANPD ordered Meta to suspend using personal data to train AI, citing (among other things) inadequate protection of children’s and adolescents’ data (ANPD). According to Brazil’s data authority, the order of July 2, 2024 was lifted on August 30, 2024 under a compliance plan in which Meta committed not to use data from accounts of people under 18 to train its AI until a final decision.
- United States. As far as we know, there is no federal opt-out from AI training on public content. For US users, the practical lever is account privacy, not a form.
The trend is consistent: regulators increasingly treat children’s data as deserving special protection. But rights you have to exercise, form by form, region by region, are a fragile kind of protection compared with simply not exposing the photo.
How to opt out, step by step
If your child’s photos are already on public accounts, here’s the realistic order of operations:
- Make the accounts private. This is the single biggest lever, and it works everywhere, including the US, where, as far as we know, there is no opt-out from Meta. A private account is out of scope for public-content training.
- File the objection, if you’re in the EU, UK, or Brazil. In Meta’s app, go to the Privacy Center and search for “right to object” to AI training, then submit the form. Do it before you post more, since objections generally apply going forward.
- Prune what’s already public. Take down old public photos of your child you’d rather not have circulating. It won’t reach datasets that already copied them, but it stops new scraping.
- Check the school and clubs. Ask whether their social pages can leave your child out. This is a normal request, covered more in our sharenting guide.
Steps 1 and 3 are really the same idea: reduce what’s public.
The fix that holds: don’t post publicly
Opt-out forms are useful, but they’re slow going: one company, one region, one setting at a time, and they do nothing about the datasets that already copied a public photo. The move that holds is upstream: don’t put the child in front of the public in the first place.
That upstream move is the whole premise of pouchie. You share your kids’ photos into a named circle instead of a feed, which means there’s simply no public surface for a scraper or a training crawler to reach. Nothing you share is used to train AI or sold to data brokers, the location and camera data is stripped from every photo before anyone opens it, and screenshots are blocked so an image can’t be captured and passed along. The AI question never really arises, because the photo was never public to begin with. If you want the step-by-step, here’s how to share with grandparents privately.
Where both answers lead
Does Meta use your child’s photos to train AI? Yes, if they’re public. Does Google mine your private Google Photos library for it? By its own statement, no (though the boundary shifts with the features you turn on, which we unpack in our Google Photos alternative for family photos). But both answers land in the same place: what makes a photo reachable is that it is public, and no single opt-out fully undoes that. You don’t need to delete your family’s digital life or live in fear of a settings menu. You need to move the photos of the people who can’t consent yet out of public view, and share them inward instead, to the small circle you chose.
