AI and copyright: the data of the problem

Copyright is an old law: it was created in 1791 with the Le Chapelier law. However, it is constantly evolving, in close symbiosis with technological innovation. The latest development that is shaking up the old law is generative artificial intelligence.

The questions that artificial intelligence poses to copyright are of two types in consideration of two categories of data:

‍

- Input data ; and

‍

- Les output data.

‍

I. Input

Remuneration of copyright owners: a win - win deal 

The observation, as it results in particular from CSPLA report on the implementation of the European regulation establishing harmonized rules on artificial intelligence (RIA), published on December 11, 2024, is as follows: quality data is essential for training AI models. Indeed, if an AI model is trained with data that is itself generated by artificial intelligence, it collapses.

It is therefore important, not only for data owners and copyright holders, but also for AI models, to start a dialogue in order to recognize the commercial value of this data.

During an open commission of lawyers “AI and copyright — how to build an ethical and competitive market” on January 16, 2025, Mr. Alexandre Lasch, Director General of SNEP, explained that, during the adaptations and negotiations relating to streaming (in its time, another technological advance) of the two operators involved, Radioblog and Blogmusik, it was indeed the one who negotiated with the copyright holders who still exists (now Deezer) and not the one who refused... (Radioblog was ordered to pay 1 million euros to collective management organizations).

Thus, the remuneration of owners of copyright and related rights for the use of their works by artificial intelligence models will allow creators to be fairly rewarded for their work and creativity, which promotes the production of quality content. And this remuneration will make it possible to offer AI developers access to rich and diversified resources.

The legal bases for such remuneration

The various applicable texts are:

- The Regulation on artificial intelligence (RIA) of 13 June 2024 ; and

- The Directive on copyright and related rights in the digital single market of 17 April 2019.

The RIA created (article 53) an obligation of transparency for providers of general-purpose AI and that of making available to the public a sufficiently detailed summary of the content used to train the general purpose AI model (a summary template was developed by the CSPLA).

The directive of 17 April 2019 introduced an exception to copyright for “text and data mining”, which allows AI providers to build their databases, as soon as content is accessible online, although it is specified that rights holders can oppose this mining of texts and data by the clause called “ofOptout ”.

The difficulty for right holders, however, is to provide proof of illicit access, the content used and the method of exercising the opt out.

‍

II. Output

The other question asked by artificial intelligence is whether a work created with the help of AI can be considered original and therefore protectable by copyright.

Under French law, a creation, to be protectable by copyright, must be the work of a human being (see the history of Painting by Lolo the donkey whose tail was a brush attached).

One A fairly similar debate had already taken place on the occasion of the birth of photography.

There is no doubt that a whole body of case law will emerge, or even legislation, as was the case at the birth of photography. While an 1866 law provided specific provisions concerning the protectability of a photograph, the 1957 law then removed these to establish a general framework for works protected by copyright.

During the debate on the protectability of a work created with the help of AI, there is no doubt that the Prompt (a prompt is an instruction or set of data provided to an AI system, which uses this information to generate responses or creations in text, image, or other form of media) will become the essential element in order toprovide proof of instructions given to the machine. Should we therefore expect that AI providers must keep the prompts?

It will then be up to the courts to carry out the delicate work of assessing the original nature of the creation developed using AI.

Balancing legitimate protection with the need to not protect just anything is vital. Without it, our old copyright may well die. And with him, the authors?

As always, the question is what world do we want to live in...

Véronique Piguet, partner lawyer at Squair

vpiguet@squairlaw.com