Meta has halted the training of its artificial intelligence models using data from Facebook and Instagram users within the European Union. The decision comes in response to rising demands from European regulatory bodies, delaying the company's AI service rollout in the region. Meta, however, will continue to utilize public social media posts from outside the EU for its neural network training.
Regulatory Pushback and Privacy Concerns
Meta's decision follows formal complaints to data protection authorities in 11 different countries, led by Ireland. These authorities instructed Meta to cease using EU user data for AI development. While privacy advocates see this as a positive step, Meta has expressed disappointment, suggesting that this move hinders European innovation and delays the region's access to AI benefits.
In a statement, Meta commented on the request from the Irish Data Protection Commission (DPC), saying, “We are disappointed by the request from the Irish Data Protection Commission (DPC) to delay training our large language models (LLMs) using public content shared by adults on Facebook and Instagram.” The statement notes that the company has been adjusting its practices based on regulatory input since March but sees itself unable to meet the current conditions while training its models sufficiently:
“We are committed to bringing Meta AI, along with the models that power it, to more people around the world, including in Europe. But, put simply, without including local information we'd only be able to offer people a second-rate experience. This means we aren't able to launch Meta AI in Europe at the moment.”
Meta, however, assures that it will continue to work with the DPC and address specific issues raised by the UK's Information Commissioner's Office (ICO).
Regulatory Engagement and Future Plans
The Irish DPC acknowledged Meta's decision, characterizing the interaction with the company as “intensive engagement.” The DPC and other European data protection authorities will continue discussions with Meta regarding this matter. The DPC wrote:
“The DPC welcomes the decision by Meta to pause its plans to train its large language model using public content shared by adults on Facebook and Instagram across the EU/EEA. This decision followed intensive engagement between the DPC and Meta. The DPC, in co-operation with its fellow EU data protection authorities, will continue to engage with Meta on this issue.”
Meta had initially planned to train its models on publicly available content from Europeans, explicitly excluding private messages and minors' posts, and allowing users to opt out. Meta argues that without EU data, its AI could struggle with regional languages and cultural nuances, leading to a suboptimal user experience for Europeans.
Ongoing Monitoring and Legal Commitments
Privacy organization noyb, which had filed multiple complaints across Europe about Meta's data practices for AI training, welcomed the development but plans to monitor the situation closely. Max Schrems, noyb's founder, pointed out in a statement that the complaint centers on Meta's legal basis for data collection, questioning the company's broad use of personal information under AI technology.
“We welcome this development, but will monitor this closely. So far there is no official change of the Meta privacy policy, which would make this commitment legally binding. The cases we filed are ongoing and will need a determination.”
“The Meta press release reads a bit like ‘collective punishment'. If any European insists on his or her rights, the entire continent will not get our shiny new products. However, Meta has every option to roll out AI based on valid consent – it just chooses not to do so.“
European regulators, however, welcomed the pause. Stephen Almond from the UK Information Commissioner's Office commented positively on Meta reflecting on user concerns.
“We are pleased that Meta has reflected on the concerns we shared from users of their service in the UK, and responded to our request to pause and review plans to use Facebook and Instagram user data to train generative AI.”
