Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Meta has warned that the EU’s approach to regulating artificial intelligence is creating the “risk” that the continent is cut off from accessing cutting-edge services, while the bloc continues its effort to rein in the power of Big Tech.
Rob Sherman, the social media group’s deputy privacy officer and vice-president of policy, confirmed a report that it had received a request from the EU’s privacy watchdog to voluntarily pause the training of its future AI models on data in the region.
He told the Financial Times this was in order to give local regulators time to “get their arms around the issue of generative AI”.
While the Facebook owner is adhering to the request, Sherman said such moves were leading to a “gap in the technologies that are available in Europe versus” the rest of the world. He added that Meta will not ship multimodal models in the EU, and with future and more advanced AI releases, “it’s likely that availability in Europe could be impacted”.
Sherman said: “If jurisdictions can’t regulate in a way that enables us to have clarity on what’s expected, then it’s going to be harder for us to offer the most advanced technologies in those places . . . it is a realistic outcome that we’re worried about.”
The comments come as Big Tech groups and start-ups are racing to commercialise AI products, while also being constrained by the EU’s wide-ranging digital rules, including its new Artificial Intelligence Act that seeks to regulate the development of the most powerful models and services.
This latest EU request in particular, Sherman said, was due to uncertainty over whether training AI models on consumer data was permissible within the EU’s General Data Protection Regulation (GDPR) rules. These rules require companies that collect or use personal data to obtain the individual’s consent and disclose their reason for doing so.
Sherman said Meta would not “be able to serve [European consumers] properly” without the ability to train on European data, as the AI would be less effective and could not respond to the “cultural concepts and contexts they need”.
Meta has already held off rolling out its Meta AI assistant, and a new version of its Ray-Ban Meta smart glasses with the assistant embedded, over regulatory and data protection fears in the EU and the UK. The assistant is now available in 22 countries, including the US, Australia and Argentina, and in new languages such as French, Spanish and Hindi.
Apple, meanwhile, has said it will not launch several features under its Apple Intelligence brand, owing to concerns related to the EU’s Digital Markets Act.
The EU said it did not comment on individual decisions of companies but added that it was “an attractive market of 450mn potential users and has always been open for business for any company that wants to provide services in the European internal market”, provided they complied with the law.
Despite halting training on future models, Meta on Tuesday released an updated version of its AI models — Llama 3.1 — which will be available in Europe and globally.
This includes a new 405bn parameter model, which Meta has dubbed the “industry’s largest and most capable open model ever created”. A higher number of parameters is often considered to deliver superior performance.
