Meta said its models need to understand the “diverse nuances and complexities that make up European communities” – referring to dialects, “hyper-local knowledge,” and a different use of humor and sarcasm
27 May 2025 (Brussels, Belgium) – – As of today, Meta will start training its artificial intelligence models with European user data, including the treasure trove of public posts and comments from Facebook and Instagram users – yes, your cringeworthy selfies, too.
Meta started rolling out its AI tech in the EU this past March.
Ireland’s privacy regulator blocked the company from feeding Europeans’ data to its AI last year. But the same regulator said last week that it was satisfied with the tech firm’s tweaks to address initial concerns around whether Meta had a solid legal justification and whether its data processing was transparent.
What changed, then? Last December, the European Data Protection Board (EDPB) published an opinion that said firms can rely on the company’s “legitimate interest” in the data as a legal basis for AI training. Still, they need to check whether it would harm people’s rights.
The EDPB’s report was “ambiguous” about legitimate interest, said Kris Shrishak, senior fellow at the Irish Council for Civil Liberties. “Meta was like: See, look at this report, it allows us to collect data and process it for training AI models,” he said, while others were unconvinced.
Not off the hook (1): Regulators and activists have been ratcheting up the legal pressure. Among their concerns: Meta could suck up sensitive or minor-related information, people didn’t think about AI when they initially shared their data, and Meta isn’t meeting the three-step check laid out by the EDPB.
EDRI Policy Adviser Itxaso Domínguez de Olazábal warned that if companies invoke “legitimate interest without meeting the legal criteria, the GDPR risks becoming meaningless.” Some have taken the EDPB December opinion as “a green light to bypass core principles.”
Also, once personal data is swallowed, it’s pretty hard to get it back, making it more difficult for people to use their right to erasure under the General Data Protection Regulation.
Not off the hook (2): Austrian privacy group Noyb – of Facebook tormentor Max Schrems – is considering an injunction at the EU level (although German courts rejected such a request last Friday).
Hamburg’s Data Protection Commissioner is preparing an emergency block on Meta’s AI use of German data. Meta responded yesterday morning to a letter of formal notice sent last week, the authority’s spokesperson said. She added that the authority “will now move quickly” to determine “next steps.”
Why do they need European data anyway? When Meta announced it would feed its AI European data, it said the models needed to understand the “diverse nuances and complexities that make up European communities,” referring to dialects, “hyper-local knowledge,” and a different use of humor and sarcasm:
“It will help them to produce better output in those languages, because a lot of these companies have primarily relied on English and a bit of other languages here and there”.
Not off the hook (3): Meta must file a report to the Irish DPC by October evaluating whether the company’s data protection safeguards and measures are adequate.
With the *final” word to Meta:
Meta spokesperson Matthew Pollard said last week that the company’s approach complies with the EDPB’s “clear guidance” and follows “extensive and ongoing engagement” with the Irish DPC. “We always comply”.
Stay tuned 🤪 😏
