AI and Copyright: EU MEPs Want to Take a Shortcut

Contents

Under what conditions is the training of AI models with copyrighted content permissible? While the question is currently occupying courts across Europe, providers have long been creating technological facts. However, the revision of the controversial DSM Directive, introduced in 2019, which enshrined an exception for AI training in copyright law, is still pending. The EU Commission's timeline foresees an assessment in 2026 of what action is needed – and then, if necessary, reform proposals will be introduced into the long EU negotiation process. At the beginning of August, external consulting firms are to supply the EU Commission with the necessary groundwork for its evaluation report – after which it would be carefully evaluated and then possible legislative changes proposed.

From the perspective of MEPs in the European Parliament, however, this is firstly far too late and secondly the current situation is not sustainable. Next week, the plenary session at the Strasbourg seat of the representative body of all 450 million EU citizens will vote on a so-called own-initiative report by German CDU legal politician Axel Voss – which contains a multitude of proposals on what needs to be done urgently now. It is about a "pragmatic solution" between AI providers and creators, says Axel Voss.

One of the essential core points of the proposal: In the future, there should be clear standards on how providers can signal what is possible with copyrighted content. The so-called text and data mining exception for copyright law only applies if the creators have not objected in a machine-readable way. "Using the TDM exception for a business model to create a competing product was never the idea of the law," says Axel Voss.

Specifications for machine readability are central

However, many of the legal disputes revolve around this point, machine readability: How, in what format, and with what signals can usage be permitted or denied? The report, which was already unanimously adopted in committee in January by politicians of all parties, with the exception of the far-right, provides clear specifications here. There should be a responsibility for everyone to label their works under what conditions they can then be used for AI training, explains Voss.

"If we were to simply abolish the text and data mining regulation, all AI models in Europe would fall into a very deep legal uncertainty," warns SPD MEP Tiemo Wölken. But where the Copyright Directive falls short, for example regarding transparency and remuneration, an additional framework is needed.

Trustee could review usage

To find out how works have been used, there needs to be a way to track the actual usage by providers like OpenAI, Anthropic, Suno, ForestLabs, or Mistral. The MEPs of the Legal Affairs Committee see both technical possibilities, such as watermarking, but also that of an independent and impartial trustee, such as the European Union Intellectual Property Office (EUIPO). This could then check the data of the AI operators for the creators without direct insight being granted.

The key to timely fair remuneration is that good collective agreements are reached, agree Social Democrat Wölken and Christian Democrat Voss. In other words: Providers and collecting societies would have to find a solution for the specific issues. However, this currently seems to be a long way off – and the AI providers have not only a lot of venture capital but also plenty of patience, as the pressure on them is manageable.

MEPs: AI cannibalization threatens news media

The pressure on creators is evident in another point: AI summaries in search engines and the search usage of AI chatbots are increasingly shifting user flows away from traditional media providers. From user suppliers, they become dead ends, with information from those who are now cut off. "Nothing less than the viability of news media is at stake," says SPD MEP Tiemo Wölken. A clarification that press performance rights also apply to AI use is just one of the demands.

Wölken also demands that the EU Commission examine with the highest priority whether chatbots and AI elements in search engines would be subject to the requirements of the Digital Markets Act and Digital Services Act, and agrees with Axel Voss on this: He sees the danger that reporting on daily events will increasingly be controlled by AI solutions in a non-transparent and possibly interest-driven manner. "We cannot afford to tinker with laws for years," warns the CDU politician, who was primarily responsible for the report.

What consequences the Voss report will actually have remains to be seen. Because unlike at the national level, the European Parliament cannot initiate laws on its own – the right of initiative lies with the EU Commission, which must then reach an agreement with Parliament and the Council of Member States. This process takes several months even in the shortest possible scenario. And could thus coincide with the regular DSM revision opportunity.

(nen)