AI training with user data: Noyb complains about Meta

"We are currently preparing to expand AI at Meta to your region". This is how an email that users of the US company's services have been receiving recently begins. The European data protection organization Noyb believes that Meta should first ask its users for permission before training AI models with the postings of European users. The association has filed complaints in eleven European countries and is calling on the authorities to initiate urgent proceedings to stop this use immediately. The data protection activists see "misuse of personal data".

"'AI at Meta' refers to all of our features and experiences that use generative AI, such as Meta AI and AI Creative Tools, as well as the models on which they are based," Meta writes in its emails to users. Meta wants to apply this technology more broadly, which is why the company is updating its privacy policy. The email also refers to the right to object.

Data from 4 billion people

According to Noyb, the company took a closer look at Meta's notification email. It became clear that the US company wanted to use personal posts, private images or data from online tracking, which it had collected since 2007, for an unspecified AI technology. This would collect information from any source and potentially share it with unknown "third parties". "Instead of asking users for their consent (opt-in), Meta claims to have a legitimate interest that overrides the right to privacy", writes Noyb.

Meta also wants to collect additional information from third parties and extract data from other online sources, writes Noyb. The only exception appears to be chats between private individuals, while communication with companies is not protected. The purpose for which the AI technology is to be used remains unclear and is incompatible with the requirements of the General Data Protection Regulation (GDPR). Meta's privacy policy would theoretically allow any purpose. "All of this is extremely worrying because it involves the personal data of around 4 billion people", emphasizes Noyb.

"Objection form a farce"

The opt-out form offered by Meta is a farce, say the data protection activists. "Theoretically, an opt-out could be implemented in such a way that it only requires one click (such as the usual 'unsubscribe' link in newsletters). Meta, on the other hand, makes it extremely complicated to object. Among other things, personal reasons must even be given." A technical analysis of the opt-out links also revealed that Meta requires a login to view an otherwise public page.

Read also

Meta AI läuft jetzt mit Llama 3

Instead of just providing an opt-out option, Meta should have asked for active consent from users before using AI under current law, explained the group founded by data protection activist Max Schrems. It had therefore filed complaints in Austria, Belgium, France, Germany, Greece, Italy, Ireland, the Netherlands, Norway, Poland, and Spain. Schrems had already given the Facebook group a fright in two lawsuits – and twice overturned data agreements between the USA and Europe before the European Court of Justice.

Meta AI – an AI assistant that acts as a chatbot and image generator – could soon be available in Germany. To teach the system its language skills, the software is trained with many texts from human users. Meta talks about "posts, photos and their captions, as well as messages that you send to an AI". The German Federation of Consumer Organizations suspects that even a post that someone has posted on Facebook that is only visible to friends or certain people can be used for AI training. Only private messages to friends, for example in Messenger, should not be used.

(anw)