Meta begins training its AI models on EU public content and Meta AI interactions

What Meta is actually starting this week

Meta says it will begin training its AI models on two categories of European data: public content shared by adults on its products — public posts and comments — and people's interactions with Meta AI, meaning the questions and queries users type into the assistant. The move follows the launch of Meta AI in the EU the prior month across Facebook, Instagram, WhatsApp, and Messenger.

Beginning this week, EU users will receive in-app and email notifications describing the data being used and linking to an objection form. Meta states it will honor objection forms already received as well as newly submitted ones, and that the form can be submitted at any time.

The boundaries Meta drew around the data

Two exclusions are explicit. Public data from EU accounts belonging to people under 18 is not used for training. And Meta says it does not use the content of private messages with friends and family — with a caveat added in a March 2026 update.

We do not use the content of your private messages with friends and family to train our AIs, unless you or someone in the chat chooses to share those messages with our AIs.Montana Labs

That parenthetical clarification is worth noting: the later edit specifies that messages people choose to share with Meta's AI features may be used for training. The line between private and trainable is therefore drawn at the moment a user routes a message into an AI feature, not at the messaging surface itself.

The regulatory path behind the resumption

Meta is candid that this is a restart, not a first attempt. It delayed training on public content last year while regulators clarified requirements. The company points to a December opinion from the European Data Protection Board, which it says affirmed that its original approach met its legal obligations, and describes ongoing engagement with Ireland's Data Protection Commission.

Meta also positions itself against peers, naming Google and OpenAI as companies that have already trained on European user data, and claiming its own approach is more transparent than many counterparts. The comparison frames the resumption as catching up to an established industry practice rather than breaking new ground.

Why the localization argument is the load-bearing claim

Meta's stated rationale is that models built for Europeans need European data — dialects, colloquialisms, hyper-local knowledge, and the distinct ways different countries use humor and sarcasm. It ties this to multi-modal capability spanning text, voice, video, and imagery, arguing that broader training data improves cultural fidelity.

For teams tracking how consumer AI is trained, the concrete signal here is the coupling of two data streams — historical public posts and live assistant queries — under a single opt-out mechanism. The queries in particular create a feedback loop: usage of Meta AI directly feeds the next round of training. That makes the objection form, and how easy it genuinely is to find and act on, the practical control point for anyone in the EU who wants out.