Meta is training AI on our children's photos
Confirmed under oath to the Australian Senate. Every public Facebook post since 2007. No opt-out for Australians.
The Senate testimony
11 September 2024. Meta's global privacy director Melinda Claybaugh appeared before the Australian Senate Select Committee on Adopting Artificial Intelligence. Under questioning from Greens Senator David Shoebridge, she confirmed that Meta scrapes every public Facebook and Instagram post by Australian adults since 2007 to train its AI models, including Llama and Meta AI.
When Shoebridge asked Claybaugh directly whether Meta had been scraping posts going back to 2007, she responded with a single word: "Correct."
"Meta has just decided that you will scrape all of the photos and all of the text from every public post on Instagram or Facebook that Australians have shared since 2007, unless there was a conscious decision to set them on private." Senator David Shoebridge, Australian Senate Select Committee on Adopting AI, 11 September 2024
Children's photos are included
Claybaugh confirmed that Meta excludes data from accounts belonging to users under 18. That sounds like a protection. It is not.
Photos of children posted on adult or organisational accounts are not excluded. A school Facebook Page is an organisational account run by adults. Every photo of every child posted on that public Page is available to Meta for AI training. The system does not distinguish between a product photo and a photo of a six-year-old at a Book Week parade.
Senator Tony Sheldon pressed this point, noting that parents who post photos of their "13-year-old daughter" or "nine-year-old son" would have no idea those images were being harvested for AI training. Claybaugh confirmed Meta does not exclude children's images from adult posts.
School Facebook Pages are public organisational accounts. Every photo of a child posted on those Pages is scraped for AI training. This is not a risk assessment. This is not a hypothetical. This was confirmed by Meta's own representative, under oath, to the Australian Senate.
An audit of every NSW government school found that more than 1,750 operate public Facebook Pages. Over 637,000 students attend these schools. These are the Pages that Meta confirmed it scrapes.
No Australian opt-out
Europeans can opt out of AI training under the General Data Protection Regulation (GDPR). Meta paused launching AI products in Europe entirely while regulatory uncertainty persisted. Australians get no such protection.
When asked why, Claybaugh was blunt:
"The specific option that we're offering in Europe is in response to a very specific legal framework." Melinda Claybaugh, Meta Global Privacy Director, Australian Senate, 11 September 2024
Translation: Meta provides privacy protections only where the law forces it. Australia's current privacy laws do not force it. So Australian families get nothing.
"Meta made it clear today that if Australia had these same laws, Australians' data would also have been protected." Senator David Shoebridge, 11 September 2024
Scale
Facebook has approximately 19 million monthly active Australian users. In NSW alone, an audit of every government school found 1,780 schools (82%) operate public Facebook Pages, serving 637,818 students. The pattern is consistent: 83% of metro schools, 81% of regional schools, 82% of outer regional schools. Nearly two decades of content from every one of those Pages has been fed into Meta's AI training systems.
Once a photo has been processed into AI model weights during training, it cannot be removed. There is no "delete" button for training data that has already been incorporated. Deleting the original post from Facebook changes nothing for models already trained. The patterns learned from a child's face persist inside the model permanently.
Regulatory response
OAIC issues guidance that AI training on personal information may breach the Privacy Act 1988.
Privacy Act amendment strengthens protections. Still no specific mechanism to stop AI training on public posts.
Children's Online Privacy Code exposure draft released. Consultation closes June 2026.
Children's Online Privacy Code registration due. Until then, the gap remains open.
Current status
As of April 2026, Australians have no opt-out for Meta's AI training on public posts. Moving a school Facebook Page to a private Group prevents scraping of future public posts by third parties, but Meta's terms of service for content posted within its own platform remain broad. The only way to fully stop Meta from using children's photos for AI training is to stop posting children's photos on Meta's platforms entirely.
A private Group removes content from public access, which stops third-party scrapers and search engine indexing. It does not stop Meta itself. But it eliminates the broadest attack surface: unrestricted public access to every photo of every child, by anyone on the internet.
- Australian Senate Select Committee on Adopting Artificial Intelligence, hearing transcript, 11 September 2024
- Meta grilled over Australian posts used to train AI – Information Age (ACS)
- Meta admits users will only get opt-out if governments force it – Fortune
- Photos of Australian kids used to train AI without consent – Information Age
- Meta confirms it scrapes Australian users' posts without opt-out – SiliconANGLE
- Facebook confirms it scrapes every Australian adult's public photos – HackerNoon
- OAIC guidance on privacy and AI, October 2024
Follow the investigation
Get notified when new evidence emerges or policy changes.
Last reviewed: April 2026