Meta Started AI LLM Training in Europe with Real User Data

Meta Platforms started training its Llama AI model on content posted in the European Union. The social media giant was cleared for the action after a German consumer group lost a last-ditch bid to block the move.

A German court last week rejected an application by the North Rhine-Westphalia Consumer Protection Agency to prevent Meta Platforms from training its AI Large Language Model based on content posted publicly on Facebook and Instagram. Wolfgang Schuldzinski, board member of the North Rhine-Westphalia Consumer Advice Centre, expressed disappointment at the court’s decision and argued there are persistent doubts over the legality of Meta's plan.

The consumer group argues Facebook and Instagram users should expressly consent to their content being used in the LLM training, with an assumption of a legitimate interest, not an adequate reason to proceed. “Given the volume of data in question, users should retain a sovereign right to participate and not simply be granted the option to object,” the group wrote, referring to users’ right to say no to the AI training.

The Agency explained that the data Meta intends to use can no longer be retrieved or deleted once employed for AI training. Meta also faced objections from Hamburg’s Commissioner for Data Protection and Freedom of Information, which noted users’ right to prevent their data from being processed, but also cautioned that information already fed into the AI model could not be removed.

The move is the culmination of more than a year of work Meta undertook with regulators to bring its AI training plan into line with various EU rules and address other concerns over user rights. It is required to produce a report detailing if the protections regulators sought have proven effective and appropriate within the next six months and face ongoing monitoring.