Video games and generative artificial intelligence: what legal protection for tomorrow's games?

A non-player character who holds a real conversation, music that adapts to each playthrough, scenery generated on the fly: generative artificial intelligence is transforming the way video games are made. As early as 2023, Ubisoft launched its “Ghostwriter” tool to write background dialogue for secondary characters. Square Enix integrated Google’s Gemini AI into Dragon Quest X. Sony even filed a patent for an AI capable of playing in the player’s place, called “Ghost Player”. More recently, Nvidia developed DLSS5, a technology that uses AI to enhance the graphics and visual realism of games in real time. For its part, Epic Games unveiled Unreal Engine 6 in June 2026, an engine that includes AI tools to speed up the creation of assets, characters and animations, with early access not expected before the end of 2027. But behind this technical progress lies a formidable legal question: when an AI takes part in creating a video game, who is the author? Can the studio still protect its game against copying? And could the AI itself be trained lawfully on other creators’ works?

The video game, a protected but complex work

Before talking about artificial intelligence, we need to understand how a “classic” video game is protected. Under French law, a video game is protected by copyright if it is original, meaning that it bears the imprint of the personality of its author or authors. This principle stems from long-standing case law that is still valid today (French Court of Cassation, plenary assembly, 7 March 1986, Atari and Williams Electronics). In practical terms, this means that a creation must result from free and personal choices, not from mere technical or automatic execution.

The French Intellectual Property Code (CPI) protects software as such in Article L. 112-2, 13°. But a video game is not just software: it also includes graphics, music, dialogue and sometimes a complete script. The French Court of Cassation resolved this difficulty in a well-known ruling (Cass. 1re civ., 25 June 2009, Cryo): the video game is now regarded as a “complex work”. This means that each element of the game is subject to the rules that apply to it: the software regime for the code, the audiovisual work regime for the cutscenes, the musical work regime for the soundtrack, and so on. Each building block of the game may therefore have a different author, governed by different rules.

This system, already complicated in practice for studios, becomes even trickier when generative AI is involved in producing one or more of these elements. Each building block of the game normally requires a contract assigning the rights of the person who created it. To that end, there is an automatic transfer of economic rights in software created by one or more employees in the course of their duties (Article L. 113-9 of the CPI). However, this transfer only applies to software: for the other components of the game (graphics, music, script), employees’ rights are not automatically transferred to the employer, unless the work qualifies as a collective work, and an express assignment remains necessary. So how can rights be assigned in an element which, legally, may have no human author at all? This is precisely the difficulty raised by generative artificial intelligence, and it forces studios to rethink how they secure ownership of their game.

Can an artificial intelligence be the “author” of a video game?

This is the central question. Copyright, as it has always been built, assumes that a work is the product of a human author. Yet an AI has no legal personality. If content is entirely generated by an AI, without any creative human input, it is therefore likely to enjoy no copyright protection at all, in France as elsewhere. This means that a competitor could freely reuse that content without the studio being able to object on copyright grounds.

This question has been answered very clearly in the United States. In Thaler v. Perlmutter, a researcher had tried to register an image created entirely by his AI, called the “Creativity Machine”, naming the machine itself as the author. On 18 March 2025, the federal Court of Appeals in Washington confirmed that US copyright law necessarily requires a human author. The Supreme Court declined to hear the case on 2 March 2026, so this solution is now well established in the United States. The report published in January 2025 by the US Copyright Office also specifies that a simple prompt (the instruction given by the user) is never enough to create copyright, since it merely conveys an idea, which is not protectable as such.

Nearly forty years ago, the United Kingdom adopted an unusual solution: section 9(3) of its copyright legislation recognises an “author” even for a work entirely generated by computer, namely the person who made the arrangements necessary for its creation. This was a unique exception in Europe. But in its report of 18 March 2026, the British government proposes to remove this provision, which it considers unclear and rarely used, while maintaining protection for works created with the assistance of an AI.

In France, the Intellectual Property Code says nothing about artificial intelligence. The task of clarifying the question was given to the Higher Council for Literary and Artistic Property (CSPLA), an advisory body to the French Ministry of Culture. In a report published on 16 July 2026, the CSPLA came down in favour of stability in existing law: there is no need to create a new specific right for creations entirely generated by AI, which simply remain unprotected. On the other hand, where genuine human direction remains perceptible in the final work (for example, an artist who selects, retouches and assembles AI-generated elements according to a specific intention), traditional copyright protection continues to apply.

For a video game studio, the practical consequence is simple: the more an element of the game results from a raw prompt, with no subsequent retouching or creative choice, the more likely it is to fall into the public domain as soon as it is created. Conversely, using AI as a mere assistance tool, framed by genuine human artistic choices, makes it possible to keep protection.

The other risk: what the AI was trained on

Protection of the end result is not the only problem. We also need to ask how the AI learned to generate images, voices or text. Generative AI models are trained on vast quantities of existing works, often scraped from the internet without their authors’ consent.

European law provides a specific exception for this type of use: the text and data mining (TDM) exception, which comes from European Directive 2019/790 and was transposed into French law in Articles L. 122-5, 10° and L. 122-5-3 of the CPI. In principle, this exception allows protected works to be used to train an AI, unless the author has expressly objected (the so-called opt-out mechanism). The European regulation on artificial intelligence (Regulation 2024/1689, known as the AI Act) also requires providers of large AI models to put in place a policy to comply with European copyright law. A report submitted to the CSPLA in December 2025 even argues that this obligation should apply to models marketed in Europe wherever they were trained, although this remains an interpretation rather than a rule expressly written into the regulation.

The European regulation on artificial intelligence will also eventually impose a form of transparency: content generated or substantially modified by an AI will have to be identifiable as such. This obligation, which is still being clarified, could in future directly concern studios that use AI for a significant part of their game.

This type of question arises very concretely in the voice acting of video game characters. Some voice actors’ contracts now provide that their voice may be used to train an AI capable of then generating synthetic voices for secondary characters. A model clause circulated in 2025 in the dubbing sector governs precisely this use: it allows AI to be used to improve sound in post-production, but expressly prohibits this technology from being used to generate content that would imitate or reproduce the actor’s voice without their specific consent. This is a good example of how practitioners in the sector, in the absence of dedicated legislation, are trying to secure these uses directly through contracts.

How can studios protect themselves today?

In the absence of specific and comprehensive legal rules, the legal security of a video game studio using generative AI depends largely on its contracts. Several precautions are currently recommended by video game law practitioners.

The first is to document precisely the use of AI at each stage of creation: which tool was used, with which prompts, and above all what share of human input there was after generation. In the event of a dispute, this traceability will make it possible to show that an element of the game is indeed the result of a human creative choice and therefore deserves copyright protection.

The second precaution concerns contracts with outside contractors (illustrators, composers, actors, freelance developers). Each contract should specify whether the use of generative AI is permitted and, if so, within what limits. A copyright assignment clause, which must be in writing under Article L. 131-3 of the CPI, should always accompany these services: without it, it is the contractor, not the studio, who remains the owner of the rights in their work.

The third precaution relates to the AI tools themselves. Before integrating a generative AI tool into its production pipeline, a studio should check that tool’s terms of use: some providers exclude any warranty as to the origin of the training data, which exposes the studio to risk if a third party were to claim rights in the generated content. A warranty and indemnity clause, borne by the AI provider or by the contractor using it, makes it possible to shift part of this risk. It is also advisable to provide for systematic human review or validation before any AI-generated content is integrated into the final game, if only to check that it does not too obviously reproduce the style or characteristic elements of an existing work, which would expose the studio to an infringement action.

Finally, more and more professionals recommend clearly informing players and business partners of the use of generative AI in a game, in order to anticipate future transparency obligations arising from the European regulation on artificial intelligence.

Conclusion

French copyright law was not designed for artificial intelligence, but it is not entirely helpless in the face of it either. The principle remains unchanged: no creative human input, no protection. This is the conclusion reached, each in their own way, by the American courts, the CSPLA in France and, no doubt soon, the British legislature, which is preparing to remove its provision on computer-generated works, unique in Europe. This international convergence shows that the question is no longer really whether an AI can be an author, but where to draw the line between a mere assistance tool, which leaves protection intact, and autonomous generation, which makes it disappear.

For video game studios, the challenge is therefore not to give up generative AI, whose uses are multiplying and some of which bring real creative added value. The challenge is to learn to use it as one tool among others, keeping genuine human artistic direction at every stage, and securing through contracts what the law does not yet fully regulate.

Bibliography

• Parliamentary mission report on the legal status of video games under copyright law, French Ministry of Culture, 30 May - 30 November 2011
• US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, 17 January 2025
• UK Government, Report on Copyright and Artificial Intelligence, 18 March 2026
• CSPLA, mission report on the status of artificial intelligence outputs (A. Bensamoun, J. Groffe-Charrier, rapporteur), 16 July 2026
• Dalila Madjid, « Jeux vidéo et droit d’auteur », Village de la Justice, 24 March 2020
• Reda Kohen, « La qualification juridique du jeu vidéo en droit d’auteur : l’articulation entre œuvre collective, œuvre de collaboration et catégories légales », kohenavocats.fr, 1 June 2026
• Hiro Avocats, « Le rapport de mission du CSPLA sur le statut des productions de l’intelligence artificielle », hiro-avocats.com, 19 August 2026
• Hiro Avocats, « Doublage de jeux vidéo et entraînement de systèmes d’intelligence artificielle », hiro-avocats.com, 21 August 2026
• SFA-CGT, model clause « Limitation de l’usage de l’IA dans le jeu vidéo », 14 March 2025

Kenza IKAOUILawyer in intellectual property and audiovisual law

https://lexludica.com/en/articles/video-games-generative-ai