Several European member states want to scale back protections in the GDPR in the name of European AI, but in doing so, they are primarily meeting the demands of large American AI companies.
Under the Irish presidency, the European Union is working on a weakening of the GDPR. This was observed by the privacy group Noyb. That organization likely rings a bell: its face is lawyer Max Schrems. The Schrems I and Schrems II rulings are named after him, in which European judges have twice overruled the Commission regarding insufficiently protected transfers of personal data to the US.
From extreme to still far-reaching
Noyb has reviewed several documents, including an extreme proposal and a slightly watered-down compromise text. The older proposal illustrates the general ambition of the camp wanting to change the rules; the compromise text shows how those changes might eventually end up in regulation.
The essence is as follows: several member states want to significantly weaken the protection of personal data when that data is used for AI-related matters. When a company demonstrates that it uses data for the development of AI, it would become possible to process personal data fairly freely. Protections remain, but weak wording ensures that AI companies can easily evade existing obligations regarding the processing of personal data.
AI legitimizes itself
“The processing of personal data in the context of the development and operation of an AI system, or an AI model, may be carried out for a legitimate interest of the data controller or a third party,” the compromise text states. This has been toned down from the first proposal, which literally stated that AI training and operation are always presumed to be a legitimate interest.
A presumed majority of EU member states are now saying that the interests of Elon Musk, Mark Zuckerberg, Google, or OpenAI in making enormous profits deserve priority over the European fundamental right to data protection.
Max Schrems, Noyb
The intention of amending the GDPR is to facilitate the development of AI within Europe. In practice, the exceptions are worded in such a way that they primarily give large AI specialists like Anthropic, Google, and OpenAI free rein. A company like Google could thus use all collected personal data, from customers and non-customers alike, for AI training, regardless of the age of the data.
Erosion of protection
User consent would no longer be required. Furthermore, there are calls to limit the right to have data deleted. AI companies would not have to comply if it is deemed too difficult. For now, this is not included in the compromise text.
The proposed amendments threaten to almost completely erode existing protections, Noyb warns: when an organization plays the AI card, GDPR protection would largely fall away.
Tailored to large American players
Schrems believes the proposed changes are tailored to the large players: “A presumed majority of EU member states are now saying that the interests of Elon Musk, Mark Zuckerberg, Google, or OpenAI in making enormous profits deserve priority over the European fundamental right to data protection.”

Noyb’s main criticism is the overly non-committal amendment of Article 88c in the extreme proposal (88 bis in the compromise proposal), in which everything related to AI technology suddenly receives preferential treatment. Companies are already allowed to process (personal) data, even for AI, if they obtain consent or have a legitimate interest. However, that consent must be specific, or the interest sufficiently significant. Now, ‘AI’ threatens to become a trump card that suddenly allows everything.
AI as a trump card
It is notable that not only training but also the operational functioning of AI systems is being given a free pass. Currently, there are insufficient provisions in the text to prevent abuse. For example, using personal data unasked to show advertisements is currently not allowed.
But what if that advertisement is shown by an AI system? In that case, the personal data is being used within the context of AI, which according to the proposals is permitted, Noyb notes. Meta, which already has little regard for EU law, will be all too happy to follow that interpretation.
Stronger competitive position (but for whom?)
Proponents of the changes refer to less strict rules for European companies. In practice, however, it is mainly American and, to a lesser extent, Chinese companies that sit on a vast mountain of European personal data. If the new rules pass, they will be the first to benefit.
Companies within the EU itself do not have the same access to the enormous volumes of data on Europeans, as Schrems also points out. That data is held by foreign tech giants. He argues that it is therefore madness to implement these changes with the ambition of supporting the European AI sector. In fact, the opposite threatens to happen.
The EU seems to be following the arguments of the large AI giants regarding the free use of data in a sort of panic reaction, in the hard-to-defend hope that European AI would receive a boost. The danger exists that already dominant American companies will become even more dominant. Furthermore, these companies have previously shown that their security is not in order.
Where is the economic added value?
Additionally, while the focus of the AI hype may be on the large builders of AI models, the economic added value is not necessarily there. It is becoming increasingly clear that companies deploy AI in practice tailored to their needs, and do not shy away from open models. Even in the US, there is a shift toward foreign ‘open weight’ models. It is therefore not the case that the economic future of the EU in the context of AI is linked to a free pass for OpenAI, Anthropic, Meta, Google, and others.
Schrems does not see the situation in a positive light and is considering another trip to the European Court: “Since the 1980s, we have promised people that large companies may hold the data, but they only have permission to use it in a limited way. Article 88 bis now opens the floodgates and allows unlimited use of all data for AI.”
Before it gets that far, negotiations must still take place. The draft text is not final. As Noyb indicates, much is at stake in the upcoming talks between European lawmakers.
