The call for strict rules was not long in coming when ChatGPT triggered a veritable hype about generative artificial intelligence (AI) in the spring. Even from Silicon Valley came the warning of the risk of "extinction" of humanity by such AI. The European Parliament promptly delivered, passing strict requirements in June for AI developers who seem to creatively create text, images and videos. Now, however, Germany, France and Italy are standing in the way. In a joint paper, they argue against a legal regulation.

Hendrik Kafsack

Business correspondent in Brussels.

  • Follow I follow

"We are against the introduction of untested laws and therefore propose to first rely on mandatory self-regulation through a code of conduct," says the internal paper distributed on Sunday. It is available to the F.A.Z. The Code of Conduct is intended to oblige generative AI producers to create a kind of model card for their AI. This is intended to ensure a minimum level of transparency and security, in line with the AI principles of the Hiroshima process initiated by the G7 countries.

Counter-proposal: Package insert for AI

The package inserts should contain "all the relevant information to understand how the model works, its capabilities and limitations." As an example, the paper mentions the intended use, the number of parameters used to train the AI, as well as information on tests on the risk of discrimination and security tests carried out by independent third parties ("red teams"). The paper explicitly does not provide for penalties for violations of the requirements. Their introduction could be considered if there were repeated violations, it only says.

With the catalogue of demands, the three states want to influence the ongoing negotiations on the EU's planned AI law. Negotiators from the European Parliament, the EU Commission and the Council of Ministers have been struggling for weeks to find a compromise. The "AI Act" is intended to establish clear rules for the use of artificial intelligence for the first time. European legislators want to set an example for the entire world.

The agreement should be in place on 6 December. Then the last "trilogue" of the negotiators is planned so far. The paper from Germany, France and Italy is likely to make this much more difficult. The EU Parliament's negotiators had already recently interrupted the talks and demonstratively left the room after Germany and France opposed the regulation of generative AI, which was previously considered agreed between the EU institutions. In parliament, the new paper was described as a "declaration of war". Other member states, including the Spanish Presidency of the Council of the European Union, which is actually obliged to remain neutral and is leading the negotiations for the Council of Ministers, are also advocating for strict EU rules for AI.

France as a driving force for soft regulation

The driving force behind the paper is France, which wants to establish itself as an AI location and not slow down its national AI champions such as Mistral. On Saturday, French Digital Minister Jean-Noël Barrot offered Sam Altman, the ex-CEO of ChatGPT developer Open AI, who has since been hired by Microsoft, to move to France – "where we are intensifying our efforts to put artificial intelligence at the service of the common good," Barrot wrote on the short message service X, formerly Twitter. The EU Parliament has also criticised the role of Cédric O, who, as a former digital minister, still pushed for the regulation of AI, but is now lobbying against it as a representative of Mistral.

In the federal government, there are differing opinions, not least between the FDP and the Greens, on how much AI needs to be regulated. However, Economics Minister Robert Habeck (Greens) has already moved away from strict conditions before the Heidelberg-based start-up Aleph Alpha recently raised $500 million from investors for the development of its language models, which are intended to compete with ChatGPT. "We don't want to regulate the technology, but the possible applications," Habeck said at the AI summit in Bletchley Park, England.

Federal Digital Minister Volker Wissing (FDP) praised the joint paper with almost exactly the same words: "We have to regulate the applications and not the technology if we want to play in the world's first AI league." Critics counter that the problem with generative AI lies precisely in the fact that it is not clear how it is used. Therefore, unlike "normal AI", which is developed for a very specific application, the regulation of these must also be independent of the application.