Meta’s AI training plans are putting public social media activity under new scrutiny in the UK and Australia. The company is moving ahead with the use of public posts and related content from Facebook and Instagram users, while regulators and lawmakers question how much control people really have over their data.
What Meta plans to use in the UK
In the UK, Meta plans to use public posts, comments, photos, and captions from Facebook and Instagram users to train its AI models in the coming months. The company announced the plan in a blog post.
Meta says it will not use private messages. It also says it will not use content from users under 18. That distinction matters because the program focuses on material that users have made public, rather than private conversations or accounts belonging to younger users.
The company argues that training on this content will help its AI systems better understand local context. Meta says the goal is for its models to "reflect British culture, history, and idiom." In plain terms, the company is saying that AI products intended for people in the UK should learn from the way people in the UK write, share, and communicate publicly online.
The ICO has not formally approved the plan
Meta has said it had a positive discussion with the UK’s data protection watchdog, the ICO. But the watchdog’s position is more cautious than a simple green light.
Stephen Almond from the ICO confirmed that Meta made changes, including an easier way for users to opt out. At the same time, he stressed that the ICO has not formally approved the program and is watching closely.
That leaves the UK rollout in a monitored space. Meta is continuing with its plan, but the watchdog is signaling that changes to the opt-out process do not amount to full approval. For users, the key practical issue is whether they understand that their public content may be used and whether the opt-out route is accessible enough to matter.
Australia has no opt-out option
The situation in Australia is more direct. At an Australian Senate hearing, Meta’s privacy officer Melinda Claybaugh admitted that the company uses public posts, images, and other data from Australian adults for AI training without letting them opt out.
Claybaugh also confirmed that Meta has used all public posts since 2007 for training, unless users set them to private. That means the boundary is not only about what people publish now. It also reaches back across public material that users may have shared long before AI training became a mainstream concern.
Accounts of minors are excluded. However, the source article notes a separate concern: photos of children posted publicly by parents are still used. That creates a gap between excluding accounts that belong to minors and excluding every public image that may include a child.
Why the public setting now matters more
The central distinction in both countries is public versus private. Meta’s approach depends on whether content is publicly visible, not whether a user originally imagined that content becoming part of AI training.
For Facebook and Instagram users, this makes account settings more consequential. Public posts, public images, captions, and comments can become training material under the programs described in the source. Private messages are excluded in the UK plan, and accounts of minors are excluded in the Australian account-based rule described at the hearing.
The source also shows how different national rules can shape different user choices. In the UK, Meta has made changes that include an easier opt-out, while the ICO continues to watch the process. In Australia, the company has admitted that adult users do not get an opt-out for public data.
Europe remains a pressure point
Meta’s plans have already run into resistance elsewhere. In the EU, the company paused AI training with user data in June at the request of the Irish data protection authority.
Meta described that pause as a "setback for European innovation." The phrase captures the company’s position: it sees user data as important for building and improving AI models, and it views restrictions as a barrier to that work.
Australian senators criticized Australia’s data protection environment as weak. Greens Senator Shoe bridge said Australians’ data would be protected if Australia had laws as strict as Europe’s.
The broader issue is not whether Meta can build AI models. It is how much say users should have when public social media activity becomes training data. The answer now appears to depend heavily on where those users live, what their platform settings allow, and how aggressively local regulators respond.