Meta to Train AI Models on Public EU Data

Meta Resumes AI Training on Public EU Data

Meta announced on Monday its intention to utilize publicly available content – including Facebook and Instagram posts and comments – for the training of its AI models within the European Union. This decision follows a previous pause implemented in response to concerns regarding data privacy and regulatory scrutiny.

The company stated that the training process, leveraging user-generated content within the EU, will commence this week. Furthermore, interactions users have with Meta AI will also contribute to the refinement of these models.

Delayed EU Launch and GDPR Compliance

This announcement arrives after a limited release of Meta AI in the EU last month, significantly later than its initial rollout in the United States and other international markets.

While AI training on user content has been ongoing in the U.S. for several years, Meta encountered resistance within the EU. This resistance stemmed from the region’s stringent privacy regulations, most notably the General Data Protection Regulation (GDPR).

The GDPR mandates a robust legal foundation for processing personal data, particularly when used for AI model training.

Regulatory Engagement and Restart of Training

In June 2024, Meta initially suspended plans to train its AI systems using EU and U.K. user data. This pause was a direct result of feedback from the Irish Data Protection Commission (DPC).

The DPC, responsible for regulating Meta within the EU, acted on behalf of multiple data protection authorities across the bloc.

Subsequently, in September 2024, Meta announced the resumption of efforts to train its AI systems using public posts originating from its U.K. user base.

Now, the company is extending this practice to include public posts from its EU user base.

EDPB Opinion and Continued Dialogue

“Last year, we postponed training our large language models using public content while regulatory requirements were clarified,” Meta explained in a recent blog post.

“We appreciate the guidance provided by the European Data Protection Board (EDPB) in December, which confirmed that our initial approach aligned with our legal obligations. We have since engaged in productive discussions with the IDPC and anticipate continuing to deliver the complete advantages of generative AI to individuals in Europe.”

User Notifications and Opt-Out Options

Beginning this week, EU users will receive notifications – both within the app and via email – detailing Meta’s intention to utilize public data and interactions with Meta AI for model training.

These notifications will include a link to a form enabling users to opt out of having their data used for this purpose. Meta assures that all previously received and future objection forms will be fully honored.

Data Privacy Safeguards

Meta emphasizes that private messages and public data from users under the age of 18 within the EU will not be used to train its AI models.

“We are committed to developing AI that is not only accessible to Europeans but also specifically tailored for them,” Meta stated.

“Therefore, it is crucial that our generative AI models are trained on a diverse range of data to accurately reflect the unique nuances and complexities of European communities. This includes regional dialects, colloquialisms, localized knowledge, and the distinct expressions of humor and sarcasm across different countries.”

Industry Alignment and Ongoing Scrutiny

Meta highlights that it is following the precedent set by companies such as Google and OpenAI, both of which have already leveraged data from European users for AI training.

However, the DPC continues to closely monitor the practices of large language model developers regarding AI training. Last week, the regulator announced an investigation into xAI’s training of Grok.