Meta Platforms Inc. plans to resume its use of UK adults’ Facebook and Instagram posts in artificial intelligence training projects.
The company disclosed the decision today. The move comes about three months after Meta paused the practice over privacy concerns.
Meta develops the popular Llama series of open-source large language models. The largest model in the lineup, Llama 3.1 405B, was touted as the most capable open-source LLM on the market at the time of its July launch. It managed to outperform OpenAI’s GPT-4o across some tasks during evaluations carried out by Meta.
The training data with which the company builds LLMs includes public user posts from Facebook and Instagram. Earlier this year, Meta stopped using such content for AI training in the European Union after regulators asked it to pause the practice. The company also extended the pause to the UK in a bid to “address specific requests” from the Information Commissioner’s Office, or ICO, a data protection regulator.
Meta said today that it will resume its use of UK adults’ social media posts for AI training in the coming months. According to the company, it has modified the way it processes that data based on feedback provided by the ICO.
Meta’s discussions with the regulator focused on the “legitimate interests” provision of the GDPR privacy law. The provision specifies that a company may process consumer data such as Facebook posts if it has strong reasons for doing so. Those reasons can include commercial interests.
To use GDPR’s legitimate interests provision as the legal basis of a data collection project, a company must show it “uses people’s data in ways they would reasonably expect and which have a minimal privacy impact.” A company can use such information in other ways as well, but only if it has a “very good reason” for doing so.
The GDPR also establishes other requirements for using the legitimate interests provision. Notably, a company must demonstrate that the data it collects is necessary for the project in which it’s used. It’s also necessary to establish that there isn’t a more privacy-friendly way of going about the task.
Before adding UK adults’ Facebook and Instagram posts to its AI training dataset, Meta will send affected users a notification about the policy change. The notification is set to include an explanation of the company’s data processing practices. Meta also plans to launch a form that will allow users to opt out of AI training.
“This means that we can bring AI at Meta products to the UK much sooner, and that our generative AI models will reflect British culture, history and idiom,” Meta stated.
The company’s use of social media posts to train AI models is still facing scrutiny in the EU, the other market where it paused the practice earlier this year. In July, Meta stated that it won’t make an upcoming multimodal version of Llama available in the bloc over the “unpredictable nature of the European regulatory environment.” Future, more capable versions of that model will likewise be inaccessible.