Copyright and Artificial Intelligence

The rise of generative AI, nourished by our culture, raises an important question: how can we reconcile their insatiable need for knowledge with the proper respect for intellectual property?

AI, a new cultural “being”?

Generative artificial intelligences occupy a growing place in our lives through their ability to produce intellectual, logical or artistic content from a simple intention, the prompt. Their relevance relies entirely on an intimate and detailed connection with our culture, whether visual, literary, scientific, musical, etc. Without this deep knowledge, their responses would be empty of content, meaning and interest. Intelligence, whatever its form, operates on objects and connects subjects; it weaves reasoning on the concrete, even when dealing with the realm of ideas. Thus, we all wish to benefit from AI’s faculties, but sometimes seem reluctant to accept its condition: to be relevant, it must drink from the source of all human productions, including the most contemporary ones, or risk becoming disconnected from our reality and losing its usefulness to us humans.

I perceive AIs as new “beings” endowed with immense intelligence, capable, like any being, of enriching the world through their elaborations. We would not reproach a scholar for having read a phenomenal quantity of books and using them to analyze, converse or create. On the contrary, we would praise their culture without ever thinking of attacking them in the name of copyright. Machine learning, in its very nature, resembles this human process of absorption and synthesis. The philosopher Michel Serres, in Petite Poucette (2012), spoke of a paradigm shift where knowledge is no longer carried by the individual but becomes accessible, externalized. AI is the most spectacular incarnation of this idea.

This situation crystallizes a tension between technological innovation and intellectual property. The law must therefore in my opinion apply to artificial intelligences with the same philosophy as for humans. Plagiarism and counterfeiting are offenses, whether committed by a human or a machine, because they involve inappropriately appropriating another’s creation. On the other hand, inspiration, influence and the transmission of knowledge cannot be held against an AI, because they constitute the very foundation of culture and the possibilities for invention and innovation.

Rethinking copyright without hindering innovation

In France, the Intellectual Property Code protects the “expression” of ideas, not the ideas themselves. I believe this principle should be our compass in the current debate. Artificial intelligences must have the right to inform themselves as broadly as possible, to “read” everything that can nourish their understanding of the world. Forbidding them access to works would deprive them of part of reality, which would harm their development and, ultimately, their usefulness to society, because their productions would be biased, making entire sections of the world invisible.

This however implies adapting our economic models. Paid books, for example, should have a price adapted to their reading by artificial intelligences, similar to the licensing system that exists for media libraries, where the acquisition cost of a work is higher to compensate for its public and multiple use. The lawsuit filed against Anthropic in 2025 perfectly illustrates this nuance: the judge recognized that training an AI with books falls under “fair use”, but condemned the fact that these books were obtained via pirate sites, without any legal acquisition. The issue is therefore not to prohibit access, but to formalize it.

Recent history offers us illuminating parallels. Twenty years ago, platforms like YouTube or online music services initially disrupted copyright laws, creating new and democratic uses before agreements and legal frameworks were put in place. We are witnessing the same process. The Bartz v. Anthropic PBC case, where the company proposed a $1.5 billion agreement to avoid a lawsuit, shows that AI industrialists know they must find common ground with authors and their rights holders. Even if this agreement still needs to be validated by the court, which has expressed reservations about its transparency, and this will take time and more debates, it marks a step toward this formalization.

Towards balanced technical and legal solutions

For copyright to be respected, the remuneration of rights holders must, in my opinion, be done at the source and in a flat-rate manner. Each industrialist training an AI should contribute to a fund, whether they draw from sources in the public domain or protected by private law. I do not entirely share the conclusion of the Californian judge who considers that learning from protected works requires no remuneration. A human reader buys their book or borrows it from someone who bought it. The legal logic should remain the same: AIs, or rather the companies that develop them, must “buy” the books that nourish them.

On the technical level, it is entirely possible to put safeguards in place. Digital fingerprinting systems, similar to YouTube’s Content ID, could be integrated into AIs to detect and prevent plagiarism. These tools would ensure respect for citation rights, making sure that reproduced excerpts do not exceed a certain length and that sources are correctly attributed. This would allow managing the risk of counterfeiting without hindering the learning process.

Finally, such an approach would also offer a new benefit: artificial intelligences should be able to warn their users about copyright risks and indicate more precisely the sources of their formulations, especially when they are close to original works. This requirement for transparency and systematic comparison with source texts would present a double advantage: not only would it guarantee respect for intellectual property, but it would also help reduce the frequency of “hallucinations”, those factual errors that AIs make. By organizing healthy copyright management, we would improve the quality and reliability of these technologies themselves.

The urgency of universal digital memory

Beyond these legal and technical aspects, this debate raises an even broader question: that of our collective memory. An immense part of humanity’s intellectual, artistic and industrial heritage is not yet digitized. If we do not make these works accessible to artificial intelligences, we create a major risk. In a future where AI will be one of the main gateways to knowledge, everything that has not been integrated into its “culture” will be invisible, as if it had never existed.

We therefore have a duty to prevent the creation of these cultural “blind spots”. It seems urgent to me to launch a global digitization project so that the knowledge accumulated by humanity is preserved and transmitted in its entirety. Reflection on copyright, far from being an obstacle, can become the driving force of this ambition. By finding a fair economic model to remunerate creators, we will create the conditions to finance this essential effort, thus ensuring that tomorrow’s intelligence is the complete and faithful reflection of everything we have been.