You Already Said Yes. (Meta, Artificial Intelligence and Your Old Posts)

Have you ever posted a photo on Facebook or Instagram and set it to "Public"? If you're an adult in Australia, Meta can use that post to train its artificial intelligence and there's no way for you to opt out. In this primer I explain which part of Meta's Terms of Service allows this, why it matters for power online, and what we can do about it.

1. What do the Terms of Service say?

When you sign up to Facebook or Instagram, you agree to Meta's Terms of Service. Most of us never read them. The key part is the licence you give Meta over everything you post:

"…you grant us a non exclusive, transferable, sub licensable, royalty free, and worldwide license to host, use, distribute, modify, run, copy, publicly perform or display, translate, and create derivative works of your content…" (Meta, 2026)

In plain English, Meta can use your posts almost any way it likes, anywhere in the world, without paying you. In 2024 Meta also updated its Privacy Policy to say it uses people's information "to develop and improve AI at Meta" (Mearian 2024). An AI model trained on your photos and captions fits easily inside "derivative works."

Then from 16 December 2025, Meta started using what people type to its Meta AI chatbot to decide which ads and posts they see (Meta 2025). There's no setting to turn this off.

2. How do we know this is happening in Australia?

In September 2024, Meta's global privacy director, Melinda Claybaugh, gave evidence to a Senate inquiry into AI. Greens Senator David Shoebridge asked whether, unless people had set their posts to private, Meta had scraped every public photo and post on Facebook and Instagram since 2007. Her answer was "Correct". She also confirmed that public photos of children posted by their parents would be included (Evans 2024).

After the inquiry, the Senate committee's final report described Big Tech's behaviour as "unprecedented theft" and recommended new laws for high risk AI (Siemienowicz 2024).

Same app, different rights

Senator Shoebridge put it simply, "There's a reason that people's privacy is protected in Europe and not in Australia, it's because European lawmakers made tough privacy laws" (Evans 2024).

3. Why does this matter?

This might sound like a privacy problem you can fix by changing your settings. I'd argue it's really about power and who gets to make the rules online.

• Terms of Service work like private laws. Suzor (2019) argues that platforms govern us through contracts they can change whenever they want, with very little accountability. Nobody voted on Meta's AI policy, but a policy update applied it to posts going back to 2007.
• Big Tech is a new kind of authority. Srivastava (2023) explains that big tech companies gain power by controlling "critical bottlenecks of knowledge, connection, and desire." Almost 20 years of Australians' posts is exactly that kind of resource.
• Clicking "I agree" isn't real consent. Stardust et al. (2023) argue that users should have a say in how their data is sold, monetised or shared. Someone who joined Facebook in 2009 couldn't have agreed to AI that didn't exist yet.
• From chats to targeting. Dobber (2023) shows that microtargeting gets around privacy protections by design, and warns that it harms democratic debate. Using private chatbot conversations for targeting makes this worse.
• Platforms only change when forced. Leaver (2021) showed how Facebook fought Australia's News Bargaining Code. The AI opt out follows the same pattern. Europeans got it because of regulators, not because Meta chose to offer it.
• Not all content is "property." Brown and Nicholas (2012) point out that Indigenous cultural knowledge can be owned collectively. Meta's licence treats every post as belonging to one individual.

4. The technical side (code as governance)

The architecture of the internet regulates behaviour as much as the law does. That applies here in three ways.

1. Default settings decide what's "public." The Terms give Meta the legal permission, but the audience setting on each post decides whether it goes into the training data. Settings people chose years ago are now being used for something they never expected.
2. There's no undo button. Meta said that making posts private now stops future use but doesn't affect data already collected (Jennings-Trace 2024). Australia's privacy regulator also warns that "model unlearning" isn't reliable yet (Otorepec 2024).
3. The rules only go one way. Meta's Terms ban users from collecting data from its platforms by automated means without permission (Meta, n.d.), yet Meta collects our data at a huge scale. When Signal tried to run Instagram ads showing people the data used to target them, Facebook shut the campaign down (Schroeder 2021).

5. What can we do? (call to action)

Walsh (2018) describes countersurveillance as ordinary people pushing back and making the powerful more visible. Alternatives like Mastodon use a shared "covenant" with users instead of a one sided contract. But changing your own settings only protects future posts. Real change needs the law.

The timing is good. On 31 August 2026 the government released a draft of the Privacy Amendment (Personal Data Protection) Bill 2026. It includes a new "fair and reasonable" test for how data is used and a right to have data erased by large platforms (Mervin 2026). The government has also already ruled out a copyright exception that would have let AI companies train on creative work for free. This shows public pressure works.

References