Meta AI: Using European user data for training

Whether OpenAI, Google, or Meta – many major tech companies now offer intelligent chatbots. Since ChatGPT kicked off the artificial intelligence hype in 2022, so-called large language models (LLMs) have found their way into all kinds of services. Be it in Google's Android operating system, where the Gemini service operates, or in services like Instagram, Facebook, and WhatsApp, which belong to the Meta group.

Since March, users in the EU have seen a button there that also leads to an intelligent chatbot. Meta's chatbot is based on the Llama model. You can ask the AI ​​questions or have it solve complex tasks directly within WhatsApp or Instagram.

But Meta's AI doesn't just work; it's trained – with massive amounts of data. This is now also relevant for people within the European Union. In the future, Mark Zuckerberg's company plans to use data from EU users to train its AI models. This applies to all kinds of things that have been publicly posted on the platform in recent years, such as posts, comments, and photos. Anyone who doesn't want this to happen has the opportunity to take action – but it must be done within the next few days.

According to the tech company itself, Meta needs the data to improve its AI models. On its information page, Meta explains: "We see it as our responsibility to develop AI that is not only available to Europeans, but is actually developed for them." Therefore, it is important that the AI ​​models are also trained using data from European users.

The company says it wants to ensure that the models "understand the incredibly diverse nuances and complexities of European communities." This includes, for example, dialects and colloquial language, as well as different styles of humor and sarcasm in different countries.

The explanation is certainly plausible: AI models are only as good as the data they are trained with. Currently, the Llama model is likely to have primarily incorporated US data—but this also affects the responses. AI experts repeatedly criticize the idea that large-language models are also little worldview machines . More diversity and cultural influences are therefore likely to make a model better per se.

As a general rule, Mark Zuckerberg's company only uses data from users of legal age for AI training – according to the company, data from people under 18 is not affected. Furthermore, no private messages, such as chats with friends or family, are included in the training. However, on an information page, the company explains that data from the use of AI stickers in chats, such as searches for a sticker, can be used for training.

Otherwise, the training data primarily concerns public posts posted by users on the platforms. Meta specifically mentions posts, photos and their captions, as well as comments. This likely includes postings that are years old and that the author may no longer remember.

Meta also explains that even users who don't actually post on the platforms may be included in its data pool. For example, if they are part of an image shared publicly by others on the platforms or if they are "mentioned in public posts or captions."

Meta's artificial intelligence has been available in services like WhatsApp and Instagram for a few weeks now. Now the company has announced that the chatbot will soon be more politically "balanced." What initially sounds harmless could be one thing above all: a new kowtowing to Donald Trump's policies.

Like all other tech companies operating in the EU, Meta is subject to strict data protection regulations. Some believe that its planned AI training using European data violates these laws. The North Rhine-Westphalia Consumer Advice Center filed a preliminary injunction against Meta in mid-May. The Cologne Higher Regional Court is expected to halt the planned use of personal data for AI training in an expedited procedure. The Consumer Advice Center had already issued a warning to the company a week earlier.

Meta sees it differently: "Our approach is in line with the clear guidelines set out in December by the European Data Protection Board and follows extensive and ongoing discussions with the Irish Data Protection Commission, our lead data protection regulator in Europe," the company said in a statement.

People in the EU have been given the opportunity to object to the use of their information for these purposes and have been informed of this right via email and in-app notifications.

Anyone who doesn't want to rely on the outcome of the procedure can object to the processing of data for Meta AI. However, it's important to act quickly: Meta has announced that it will begin AI training on May 27. This means that objections must be filed by May 26 at the latest.

Specifically, the objection is made via a form where you only need to enter your email address. There is also a field where you can explain your decision – but this is optional. Once you have completed the form, Meta will send a confirmation to the provided email address. In it, the company confirms that it will "not use the data for the future development and improvement of generative AI models for AI at Meta."

Different forms must be completed for Instagram and Facebook. You can find the form for Facebook at this link , and the form for Instagram at this link . Important: The forms will only be displayed if you are logged into your account. This is necessary so that Meta can assign your objection to your account. There is no form for WhatsApp, as the encrypted chats are not used for AI training.

If the above-mentioned links don't work, you can also opt out on your smartphone using the respective apps for the services. Here's how:

Facebook:

• In the app, click the icon with the three lines. This is located either in the top right or bottom right corner (depending on the app version and operating system). For some users, a small profile picture appears there instead.
• Then click on the “Settings and Privacy” tab.
• There you'll find a button called "Privacy Center." Click it.
• The following is an information text about meta-AI. In the second paragraph of this text, there's a word with a link: "object." Click on it – and you'll be taken to the objection form.

Instagram:

• Open your profile page and click the button with the three lines in the top right corner – this will open the settings.
• Scroll down until you see the "Privacy Center" button. Click it.
• Similar to Facebook, an information text about the meta-AI appears. In the second paragraph of the text, the word "object" is accompanied by a link. Clicking on it will open the objection form.

Note: Depending on the app version and operating system, the app settings may look slightly different.