Out-Law News Lesedauer: 3 Min.

European Council prepares for debate on copyright under AI Act


After an internal meeting of the European Council’s working party on IP focusing on artificial intelligence and copyright, an expert has called for an international agreement on the text and data mining exception.

The meeting took place last week was intended to offer the chance to exchange views on copyright and AI and the possible need for regulatory action.

According to a leaked document the discussion focused on the use of copyright-protected content for the training of AI models, the opportunity of IP protection of copyright-generated content, watermarking of AI-generated content, especially when there is a significant human input, remuneration schemes for rightsholders and a specific liability regime for infringements in the context of Gen AI.

The meeting took place at a time when the newly established European AI Office is preparing to write codes of practice for general purpose AI (GPAI) models. The European AI Office will play an important role in implementing the AI Act.

The EU's AI Act is expected to be published in the Official Journal on 12 July and will enter into force 20 days later. This will mark the beginning of a phased implementation: Within three years, the various rules and obligations of the AI Act will become mandatory.

With regard to the envisaged codes of practice, MEPs are warning against an over-reliance on model developers for input, to the disadvantage of other stakeholders such as civil society organisations, experts and copyright holders.

Maureen Daly, a Dublin-based IP expert at Pinsent Masons, commented: "The text adopted by the European Parliament contains provisions aimed at creating a level playing field regarding AI training data. Training AI systems requires the use of large amounts of content, some of which may be copyright protected works. There is a text and data mining, exception under EU copyright law which AI developers would seek to rely on to justify the extraction and reproduction of copyright protected works but they can only do this if rightsholders have not opted out. The opt-out requirements under the TDM is not simple and leaves open many questions on how it operates in practice. It would be great if the EU addressed those questions rather than wait for clarity to be provided by the courts as that takes time."

Article 3 of the EU DSM Directive provides a text and data mining (TDM) exception for reproduction and extraction made by research organisations and cultural institutions, and Article 4 provides a TDM exception for everyone, although with an opt-out option for rightsholders.

After the launch of ChatGPT, it was discussed whsether Article 4 of the DSM Directive covered the training of AI models with works from the internet, as the works are always reproduced at least temporarily. Much of the debate centred on how to check whether the AI has complied with the limits of the exception in Article 4.

"The EU should consider now whether TDM needs to be broadened and should this happen, it would then need to consider whether compensation should be provided to rightsholders," Daly said.

The EU AI Act does not contain any new provisions on exceptions to copyright. However, recital 105 states that the use of copyright-protected content requires the authorisation of the rightsholder, unless an exception applies. The text and data mining exceptions of the DSM Directive are generally applicable. However, if rightsholders have opted out from TDM, AI model providers must obtain their authorisation for the use of such protected content for text and data mining purposes.

Daly said "Recital 106 of the EU AI Act obliges providers to put in place a policy to comply with 'Union law on copyright and related rights' in order to create a level playing field among providers of general-purpose AI models. That way, no provider will be able to gain a competitive advantage by applying lower copyright standards than those provided in the EU."

In order to make compliance with these rules verifiable, providers of AI models must fulfil the documentation obligations under Article 53 of the AI Act and Annex XI. This means that they must document what information the AI has used for testing and training purposes and how it it was obtained. The documentation must always be kept up to date and made available to the AI Office on request.

To address the concerns of authors, the AI Act contains two further specific requirements that providers must fulfil. According to Article 53, they must introduce and operate a copyright policy to comply with EU law and in particular to identify and comply with (including through state of the art technologies) the opt-out reservations of authors pursuant to Article 4 of the DSM Directive. Additionally, Article 53 of the AI Act obliges providers to "draw up and make publicly available a sufficiently detailed summary about the content used for training of the general-purpose AI model" - with the aim of ensuring a uniform standard for this, the AI Office is about to prepare a template for such a summary.

"There are differences between the EU, US and UK on how they approach TDM. An international policy or agreement on TDM would help legislators and courts create that level playing field referred to in Recital 106 of the EU AI Act" Daly said. 

We are working towards submitting your application. Thank you for your patience. An unknown error occurred, please input and try again.