Meta locks the EU out of its future multimodal AI models

European AI users are about to hit a major technological wall, and this time, even open-source software won't save them.

Meta is withholding its next multimodal AI model, as well as future multimodal releases, from customers and companies in the European Union. The tech giant told Axios that the decision comes down to the sheer unpredictability of Europe's regulatory environment.

While Meta still plans to release a larger, text-only version of Llama 3 in the EU, its cutting-edge models capable of reasoning across text, audio, images, and video simply will not cross the EU border.

2 billion opt-out notices weren't enough

The standoff isn't over the EU's newly minted AI Act, but rather its long-standing privacy framework, GDPR.

Back in May, Meta announced it would train future models using publicly available Facebook and Instagram posts from European users. To play by the rules, Meta sent out more than 2 billion notifications explaining the move and offering users a way to opt out, aiming to start training in June.

Meta claimed it had briefed regulators months in advance and addressed minimal feedback. But right after the public announcement, privacy regulators cracked down.

In June, European watchdogs, led by the Irish Data Protection Commission, ordered Meta to pause training immediately, following up with dozens of detailed inquiries.

The real intrigue? Meta plans to roll out the multimodal models in the UK, where data protection laws are virtually identical to GDPR. A Meta representative said European regulators are simply taking far longer to interpret existing rules than counterparts in other regions. Meta also pointed out that rivals like OpenAI and Google are already training on European data.

An open-source wall around Europe

What makes this move unprecedented is that Llama models are typically offered under an open license.

Meta made it clear that even though the new multimodal models will be open, European businesses won't be allowed to touch them. Even companies outside the EU could be barred from offering services in Europe if those products rely on the new models.

This creates an immediate headache for Meta's own hardware ambitions. Multimodal AI is the brain behind products like the Meta Ray-Ban smart glasses, enabling them to analyze what a wearer sees and hears in real time. European consumers will simply be cut out of that future.

Why it matters

This marks a dramatic shift in how American tech giants deal with European regulators.

For years, Big Tech followed a simple playbook in Europe: launch first, fight the fines later. Now, companies like Meta and Apple are opting to withhold their best products entirely. We think this is a deliberate tactic to turn public frustration and commercial FOMO into political leverage against regulators in Brussels.

For European builders and creators, the consequences are brutal. Cutting off open-weight multimodal models means local startups will fall behind global peers, and foreign apps powered by these models might avoid Europe altogether. Meta also rightly notes that without European data, future AI systems will struggle with regional languages, slang, and cultural context.

Meta is clearly taking a gamble here. Squeezing consent out of billions of notification pings was always aggressive under European privacy norms. But by walking away, Meta is forcing Europe to confront the actual cost of its regulatory friction: being left behind in the multimodal AI race.