EFJ calls on EU Member States to reach agreement on the AI Act
As the Council of the EU is expected to vote on the AI Act on 2 February 2024, the European Federation of Journalists (EFJ) urges the EU Member States to reach a long-awaited political agreement on this important text for journalism.
The media reported that Germany and France were still blocking several provisions just a few days before the vote, suggesting that negotiations are not yet completely concluded. They fear that this regulation will subject their startups to increased scrutiny related to foundation models.
In Germany, the coalition government could withhold its backing for the law. With an internal consultation with the ministries due to take place in the next day, the German Federation of Journalists (DJV) is calling on the German government to support the AI Act.
“We urgently need the regulation of Artificial Intelligence at the European level. The AI Act is far from perfect, and we would have liked to have achieved more. But this regulation is better than nothing,” says the DJV Chairman Mika Beuster. “The German government must not miss this opportunity now.”
In particular, transparency obligations laid down in the AI Act are essential for authors. Many companies will use journalistic content for the training of their AI models and therefore, the redistribution of revenue must be regulated not to leave it entirely at the discretion of platforms and AI companies. The European media and creative sectors, representing several hundred thousands of workers, has repeatedly called for meaningful transparency obligations.
“While we would have liked to see stronger provisions at the European level, the AI Act sets out some essential transparency obligations in particular for high-risk models. AI is already widely used in European newsrooms so we can’t wait any longer to introduce some regulations,“ said EFJ Director Renate Schroeder.
In the meantime, the EFJ is calling for the regulation to contain much stricter requirements for transparent documentation of content used for AI training. The current provisions on transparency and copyright are too weak to be applied effectively.
