Meta Plans to Train AI Model on European Data Amid Privacy Concerns

Meta announced on Monday its intention to use data from privacy-conscious European users to train its artificial intelligence models, amid ongoing concerns about data protection.

Meta announced on Monday its plan to use data from privacy-conscious European users to train its artificial intelligence models. This move comes as Meta faces data protection concerns and competes with rivals like OpenAI and Google.

Meta, the parent company of Facebook, Instagram, and WhatsApp, aims to use public data from European users to train its Llama AI large language model. The goal is to better reflect the "languages, geography, and cultural references" of its European user base.

However, Meta's efforts are hindered by strict European Union data privacy laws, which grant individuals control over their personal information. Last week, the Vienna-based group NOYB, led by activist Max Schrems, filed complaints with 11 national privacy watchdogs to halt Meta’s AI training plans before the next generation of Llama begins.

AI language models rely on extensive data to improve their predictive capabilities. Meta's AI assistant feature is integrated into Facebook, Instagram, and WhatsApp for users in the U.S. and 13 other countries, but not yet in Europe.

Stefano Fratta, Meta's global engagement director of privacy policy, stated in a blog post that without training on public European content, AI models will fail to understand regional languages, cultures, and trending social media topics accurately. Fratta emphasized that Europeans would be disadvantaged by AI models uninformed by Europe's cultural, social, and historical context.

He noted that other companies, including Google and OpenAI, have already trained on European data. Meta will exclude private messages and content from users under 18. Since May 22, Meta has notified 2 billion European users about its plans and provided an opt-out option via an online form. Meta's updated privacy policy, effective June 26, suggests that training for the next AI model will commence shortly thereafter.