In an update to a recent blog post about its drive to build AI technology for Europeans in "a transparent and responsible way," Meta revealed that it was "disappointed" by a request from the Irish Data Protection Commission, which is its lead privacy regulator in the European Union, to delay training its large language models using images, posts and other public content shared by adults on Facebook and Instagram.
Meta, which is the parent company of both platforms, stressed that it remains "highly confident that our approach complies with European laws and regulations," noting that it has been keeping data protection authorities across the EU updated about these efforts since March and has incorporated their feedback during its development process.
"This is a step backwards for European innovation, competition in AI development and further delays bringing the benefits of AI to people in Europe," the company said in its post.
Meta added that using public data to train AI models "is not unique to our services, and we're more transparent than many of our industry counterparts." The company confirmed that it's still "committed" to bringing its AI models to the EU. However, it reiterated that without including the type of "local information" that the Irish commissioner has raised concerns with, "we'd only be able to offer people a second-rate experience."
"We will continue to work collaboratively with the [Irish data protection commissioner] so that people in Europe have access to — and are properly served by — the same level of AI innovation as the rest of the world," the company said, adding that the delay will also give it a chance to "address specific requests we have received from the Information Commissioner's Office, our U.K. regulator, ahead of starting the training."
Ireland's data protection commission on Friday "welcomed" the pause on Meta's plans to train its large language model using publicly available information from Facebook and Instagram users across the EU.
"This decision followed intensive engagement between the DPC and Meta," the agency added. "The DPC, in co-operation with its fellow EU data protection authorities, will continue to engage with Meta on this issue."
The delay also won praise from Austrian advocacy group NOYB, which on June 6 lodged complaints with the data protection authorities in Ireland, Austria, France, Italy and a half dozen other European countries challenging Meta's plans to deploy its users' data to train artificial intelligence models. NOYB also joined Norway's Consumer Council in filing a grievance with the nation's data protection authority on the same day.
The groups argued that Meta's plans violate the European Union's General Data Protection Regulation, especially in light of the company's "cumbersome" opt-out process for Facebook and Instagram users, and that the national authorities have the tools to stop Meta in its tracks if they deem the tech giant's practices illegal.
In a statement Friday, NOYB Chair Max Schrems said that while the group welcomed the development, which it deemed a "preliminary" victory, it would continue to "monitor this closely."
"So far there is no official change of the Meta privacy policy, which would make this commitment legally binding," Schrems noted. "The cases we filed are ongoing and will need a determination."
The advocacy group added that the Irish authority's current stance appears to represent a "U-turn" on its part, given that the regulator had initially approved the introduction of Meta AI in the EU but seems to have "changed its mind" as the complaints have flooded in and other regulators have "pushed back in the past days."
Stefano Fratta, global engagement director of Meta's privacy policy, laid out the company's plans for deploying AI services in the EU in a June 10 blog post, which the company updated Friday with the news about its decision to delay these efforts.
Fratta said that Meta was planning sometime this year to expand to the EU its generative AI features and experiences that it already offers in other parts of the world, including its open-source large language model known as Llama and its free Meta AI assistant.
To do so, the models that power this technology would need to be "trained on relevant information that reflects the diverse languages, geography and cultural references of the people in Europe who will use them," according to Fratta.
"To do this, we want to train our large language models that power AI features using the content that people in the EU have chosen to share publicly on Meta's products and services," Fratta said. "If we don't train our models on the public content that Europeans share on our services and others, such as public posts or comments, then models and the AI features they power won't accurately understand important regional languages, cultures or trending topics on social media."
Fratta stressed that Meta is not the first company to go down this path, noting that both Google and OpenAI "have already used data from Europeans to train AI," and pledged that the company would make it a priority to develop this technology "responsibly and transparently."
"Our approach is more transparent and offers easier controls than many of our industry counterparts already training their models on similar publicly available information," Fratta added.
--Additional reporting by Jamie Lennox. Editing by Jay Jackson Jr.
For a reprint of this article, please contact reprints@law360.com.
