On Monday, Meta announced that it will be training AI models in the EU for public content such as Facebook and Instagram posts and comments. Previously after halting plans in response to regulatory pressures due to data privacy concerns. The company said this week it will begin training AI on user content in the EU. User interaction with meta AI is also used to train the model.
The announcement comes after a limited version of Meta AI, which was launched in the EU last month after its debut in the US and other global markets.
META has been training AI on user-generated content in the US for many years, but faces resistance in the EU as BLOC's strict privacy laws, particularly the general data protection regulations (GDPR), requires a clear legal basis for processing personal data to train AI models.
Meta said in June 2024 it would suspend plans to start training AI systems using EU and UK user data following a pushback from the Ireland Data Protection Commission (DPC). The DPC regulated the EU's META and acted on behalf of several data protection authorities across the block. In September 2024, META said it was resuming efforts to train AI systems using public posts from the UK user base.
Fast forward to today Meta announced that it would also do so in public posts from the EU user base.
“Last year, regulators delayed training large-scale language models using public content while clarifying legal requirements,” Meta said in a blog post. “We welcome the opinions provided by EDPB in December. This confirmed that the original approach meets legal obligations. Since then, we are looking forward to being constructively involved with IDPC and bringing the full benefits of generated AI to the people of Europe.”
Starting this week, EU users will begin receiving in-app and email notifications to show that Meta will start using public data with Meta AI to train their models. These notifications include links to forms that allow users to opt out of the data they are using. Meta says it respects all objections it has already received and what has been submitted.
Meta points out that it does not use private messages or public data from users under the age of 18 in the EU to train the model.
“We believe that we have a responsibility to build AI that is not only available to Europeans, but is actually built for them,” Meta says. “That's why it's so important that our generative AI models are trained with a variety of data, allowing us to understand the incredibly diverse nuances and complexities that make up our European community. It means that it all comes from dialect and colloquialism, from the use of moisture and irony in our products.”
Meta follows examples set up by companies like Google and Openai, both say they already use data from European users to train AI models.
Meanwhile, DPC is not fully operational from scrutinizing the extent to which Creator scrutinizes the large language model of the language model that trains AI services. Last week, regulators announced they were investigating Grok's training at Xai.