Meta's latest announcement follows the limited launch of Meta AI in the EU last month, which came well after its initial rollout in the U.S. and other international markets.
While Meta has long used user-generated content in the U.S. to train its AI, it encountered resistance in the EU due to the region’s stringent privacy regulations—particularly the General Data Protection Regulation (GDPR), which requires a clear legal foundation for processing personal data for AI training.
In June 2024, Meta said it would halt plans to use user data from the EU and U.K. for training its AI systems, after pushback from the Irish Data Protection Commission (DPC), which oversees Meta’s compliance in the EU and represents multiple data protection authorities across the bloc. By September 2024, Meta had resumed efforts to use public posts from U.K. users for training, and now the company says it will do the same with public posts from users in the EU.
“Last year, we delayed training our large language models using public content while regulators clarified legal requirements,” Meta wrote in a blog post. “We welcome the EDPB’s December opinion affirming that our original approach met legal standards. Since then, we’ve worked closely with the IDPC and are committed to bringing the benefits of generative AI to people in Europe.”
Starting this week, EU users will begin receiving notifications—via app and email—explaining that Meta will use their public content and interactions with Meta AI for model training. These notifications will include a link to an opt-out form. Meta says it will respect all previously submitted opt-outs as well as any new ones.
The company clarified that it will not use private messages or public content from EU users under 18 in its training data.
“We believe we have a responsibility to build AI that’s not just available to Europeans, but actually designed for them,” Meta said. “That means training our models on a diverse range of data so they can reflect the rich cultural and linguistic variety of European communities—from local dialects and expressions to the distinct ways humor and sarcasm are used across the region.”
Meta also pointed out that its approach aligns with industry practices, noting that other companies like Google and OpenAI have already used European user data to train their AI models.
Meanwhile, regulatory scrutiny isn’t going away. Just last week, the DPC announced an investigation into xAI's use of data to train its Grok AI model.