Facebook and Instagram are using your posts to train generative AI

New Zealand consumers have barely been alerted to this practice. It isn’t even clear when Meta’s privacy policy changed to allow data scraping for AI training purposes. Regardless, the policy is so long that it’s unreasonable to expect the average consumer to notice.

The only reason we really know about Meta’s AI training is because of Europe’s data protection laws. In Europe, customers have greater rights to refuse permission for a company to use their data. At the beginning of June, European Facebook and Instagram users were informed their data would be fair game for AI training if they didn’t opt out by June 26.

However, because there are weaker data protection laws in New Zealand, Meta has likely been training its AI on posts from New Zealand users since September 2023 or earlier. We can’t know for sure because there’s no requirement to inform us.

What is the data being used for?

Generative AI aims to imitate or predict human language or art. While different models are based on different computational techniques, the universal idea is to compose a large enough set of data to “train” the model on, so it can make better predictions.

When given a prompt, generative AI uses statistical analysis to guess what the next word or pixel should be, based on what’s in its training data. If your Instagram photos or Facebook interactions are in that training set, they’ll feed into how Meta’s AI responds in the future.

The stakes are arguably higher if you use social media for work – for example, an illustrator advertising their art on Instagram essentially has their work plagiarised every time someone types the right prompt into Meta’s image generation models.

Regulation has worked in Europe

Not only were European consumers the only group to be given a heads up about the privacy policy change – because Meta was legally required to do so – but the firm backlash from European watchdogs and regulators has forced Meta to postpone and rethink the change in European countries.

However, there’s no indication that the result in Europe will change anything about how Meta operates in the rest of the world – which is why we also need stronger data regulation to protect New Zealanders.

What you can do about it

Meta has long been in the business of collecting its users’ information. If you engage with Facebook or Instagram, it’s important to understand that Meta is harvesting as much data as it can and will use your data in whichever way will make the most money.

There’s little that we, the consumers, can do in the face of such a tech giant, but here are three ideas.

1. Demand that New Zealand’s Privacy Act is brought in line with Europe’s General Data Protection Regulation (GDPR), so we are alerted to this sort of thing in future and can refuse consent.
2. Consider using a tool like Glaze or Nightshade before uploading images to a Meta network. These tools protect against AI mimicry by adding a layer of minor changes that are invisible to humans but confuse AI algorithms.
3. Delete your Meta accounts – it’s the only way to opt out in New Zealand. You might want to check out Cara, an up-and-coming Instagram alternative with an explicitly anti-AI stance.

Unfortunately, it isn’t just Meta

Facebook and Instagram are just two platforms of many that are harvesting personal content to inform generative models.

• OpenAI developed the GPT model, the basis of both Microsoft Copilot and ChatGPT, which will be integrated into Apple’s new software on iPhone, iPad and Mac. The GPT-4 model was trained on over a million hours of YouTube videos, ignoring the copyrights owned by the videos’ creators.
• Google, which owns YouTube, reportedly didn’t put a stop to the data theft by OpenAI because it was doing the same thing to train its own AI models and didn’t want to draw bad publicity.
• Reddit, another social network, makes hundreds of millions of dollars by selling posts and comments to AI companies for training purposes.
• X (formerly Twitter) takes posts from the platform to train its chatbot for paying subscribers.

It’s increasingly difficult to be a digital citizen without having your voice and intellectual property hijacked and aggregated into a derivative slurry. Until regulation catches up and copyright laws are enforced, this is our future.