For most of the past two decades, the photos, status updates and captions that Australians posted publicly on Facebook and Instagram have sat on Meta’s servers doing a quiet second job. They have become raw material. The company now openly acknowledges that it feeds public content from adult accounts into the systems that power its generative artificial intelligence, the same technology behind Meta AI, its chatbots and the image tools baked into its apps.
What has changed, and what prompted a fresh wave of local coverage, is that Australians are at last being given a mechanism to push back. As the ABC reported, the practice has run largely unchecked here, and the ability to object has finally reached Australian users rather than being reserved for the privileged markets where privacy law forced Meta’s hand.
How we got here
The story has its roots in a Senate inquiry in 2024, when Meta’s global privacy director, Melinda Claybaugh, was pressed by Australian senators on exactly what the company harvests. Under questioning she conceded that Meta scrapes the public posts of Australian adults, and that unless an account had been set to private, content stretching back to 2007 was fair game for training. The admission landed awkwardly because, at the same time, Meta had paused its AI training plans in the European Union and offered Europeans a clear way to opt out.
The difference comes down to law. The EU’s General Data Protection Regulation gives residents a statutory right to object to their personal data being processed, and European regulators leaned on Meta hard enough that the company built an objection form and honoured it. Australia, with a Privacy Act written in 1988 and only partially modernised since, offered no equivalent lever. So Australians got the data collection without the off switch, a gap that privacy advocates have spent the past year describing as a two-tier system where your rights depend on your postcode.
What is actually on offer
The opt-out that has now surfaced for local users mirrors the European model in form, if not entirely in spirit. Through the privacy settings inside Facebook and Instagram, users can lodge an objection to their public information being used for AI training. The process asks you to submit a request, and in some cases to explain your reasoning, rather than offering a single toggle, which critics say introduces friction by design.
There are also important limits worth spelling out. The objection covers content you have posted publicly going forward and, where granted, your historical public posts. It does not claw back anything already absorbed into a trained model, because once data has shaped the weights of a large language model there is no clean way to extract it. Private messages and content shared only with friends have never been in scope, according to Meta, and posts made by other people that happen to mention you or show your face sit in a murkier category that the opt-out does not fully resolve.
Two ways of reading it
Privacy campaigners see a company doing the minimum required to look cooperative. Their argument is that an opt-out, especially one buried in settings and wrapped in a justification step, places the entire burden on the individual. The vast majority of users will never find it, never use it, and will therefore be treated as having consented by inertia. A genuine consent model, they contend, would be opt-in by default, with training switched off until a person actively agrees.
Meta’s framing is almost the mirror image. The company has consistently argued that training on public posts is both lawful and necessary to build AI that understands Australian language, culture and context rather than defaulting to an American worldview. In its telling, information a person chose to publish to the open internet carries a lower expectation of privacy, and an objection process is a reasonable accommodation rather than an admission of wrongdoing. Executives have also pointed out that rivals including Google and OpenAI train on vast swathes of public web data with far less transparency about which specific users are involved.
The Australian stakes
For Australia, the episode is less about one company and more about the regulatory vacuum it exposes. The federal government has been grinding through a long-promised overhaul of the Privacy Act, and reforms passed in late 2024 introduced a statutory tort for serious invasions of privacy along with new transparency rules for automated decision-making. Yet the broader question of whether Australians deserve a GDPR-style right to object to data processing remains unfinished business, and it is precisely that missing right which left locals a step behind Europeans on AI training.
There is a competitive dimension too. Australia’s own AI builders, from health startups to the sovereign-capability push gathering pace across the sector, are watching how regulators treat training data. If the rules remain loose, global giants with enormous existing data hoards keep their advantage. If they tighten in a way that only captures the small players, the effect could be the opposite of what policymakers intend. The country is trying to grow a homegrown AI industry while also protecting citizens from the data appetites of the largest platforms, and those two goals do not always point in the same direction.
The human stakes are plain enough. Millions of Australians have spent years documenting their lives on Meta’s platforms, from wedding albums to small-business storefronts to community group chatter. Most never imagined that material would one day help teach a machine to write, draw and converse. The opt-out, however imperfect, is the first acknowledgement that they should at least get a say.
What is next
Expect the objection mechanism to become a live test of how many Australians actually engage with it, and of whether the Office of the Australian Information Commissioner takes a view on whether the process meets the spirit of consent. The next tranche of Privacy Act reform, still working its way through Canberra, will determine whether the right to object becomes a legislated entitlement rather than a discretionary feature a platform can withdraw. And with every major AI developer now hungry for fresh training data, Meta is unlikely to be the last company whose data practices land in front of a Senate committee.
For now, the practical advice is unglamorous. If you would rather your public posts stayed out of Meta’s AI, the setting exists, but you will have to go looking for it.
Sources: ABC News.
