Meta Revives Initiative to Train AI with Public Facebook and Instagram Posts from UK Users

Meta has announced the resumption of its efforts to train AI systems using publicly available posts from its U.K. user base on Facebook and Instagram. The company claims to have implemented feedback from regulators, introducing a revised "opt-out" approach aimed at enhancing transparency regarding its data practices. This initiative is framed as a means for its generative AI models to better represent British culture, history, and language. However, the specific changes in this latest data collection effort remain unclear.

Starting next week, U.K. users will receive in-app notifications detailing Meta's plans. The company is set to utilize public content for AI training in the upcoming months, specifically on data where users have not explicitly opted out through the designated process. This announcement follows a three-month pause in Meta's plans against a backdrop of regulatory scrutiny in the U.K. The Information Commissioner’s Office (ICO) expressed concerns about how user data would be utilized in training generative AI algorithms and how consent was being obtained. Additionally, the Irish Data Protection Commission, which oversees Meta’s privacy operations in the European Union (EU), raised objections after consultations with various data protection authorities across the region. It remains uncertain when, or if, Meta will reactivate AI training in the EU.

To provide context, Meta has previously expanded its AI training using user-generated content in markets like the U.S. However, Europe’s stringent privacy regulations have posed challenges for the company and others trying to expand their data training efforts. Earlier this year, Meta informed EU users about a forthcoming privacy policy change that would allow the use of posts, comments, and photos for AI training, justifying this by stating a need to reflect the diverse languages and cultural nuances of its European audience.

Though intended to take effect on June 26, Meta’s announcement prompted the privacy rights organization noyb ("none of your business") to lodge multiple complaints with EU member states. They argued that Meta’s practices violated the General Data Protection Regulation (GDPR) by employing an opt-in mechanism that undermined users' ability to provide consent upfront. Instead of allowing users to opt in, they must actively object to the use of their data. Meta contends that it's relying on the GDPR's “legitimate interest” legal basis, claiming its actions comply with regulations despite skepticism from privacy experts regarding the appropriateness of this basis for data use.

Meta has previously attempted to leverage this legal rationale for processing users' information in micro-targeted advertising. However, a ruling from the Court of Justice of the European Union last year established that this legal basis could not be used in that context, casting doubt on its current AI training endeavors.

The decision to focus its efforts in the U.K.—rather than the EU—highlights the regulatory environment's differences post-Brexit. Although U.K. data protection law is still aligned with GDPR, the ICO's regulatory enforcement power is now distinct from the EU framework, and U.K. lawmakers are exploring potential deregulations of privacy laws.

Opt-out Objections

A major point of contention regarding Meta's initial approach was the convoluted process for Facebook and Instagram users to opt-out of having their information used for AI training.

Instead of offering a straightforward opt-in or opt-out option, the company made users navigate through multiple steps to find a hidden objection form. Users were required to specify their reasons for opting out, with Meta retaining discretion on whether to honor these requests, despite public claims of compliance.

In the upcoming rollout, Meta will maintain the objection form structure, necessitating users formally to notify the company if they wish to exclude their data from AI training. Those who previously objected will not need to resubmit their requests. Meta claims to have simplified the objection process based on ICO feedback, though specific details on these changes have yet to be disclosed.

Stephen Almond, ICO Director of Technology and Innovation, emphasized that the organization will "monitor the situation" as Meta proceeds with its plans to utilize U.K. data for AI model training. He stated, “It is for Meta to ensure and demonstrate ongoing compliance with data protection law. We have made it clear that organizations using user information for training generative AI models must be transparent about data utilization. They should adhere to our guidance and establish effective safeguards prior to processing personal data for model training, including providing a clear, simple route for users to object to such processing.”