
Summer Newsletter

✨AI ✨
➡️ Deepfakes and AI scams: the Danish government is about to grant its citizens copyright on our voice, our image, our identity!
In a context of rapid growth in artificial intelligence, Denmark is set to become the first European country to recognise copyright on the image, voice and physical appearance of each individual.
🔵The stated objective : to fight against the proliferation of deepfakes, particularly of a sexual or defamatory nature.
🔵Concretely:
- According to the law, it would allow anyone to request the removal of “deepfake” content from online platforms and to claim compensation.
- Platforms that do not cooperate would face heavy fines.
- In the event of non-compliance, the Danish authorities could refer the matter to the European Commission, which has the power to impose sanctions on the basis of European legislation.
- However, parodies and satires would remain allowed in order to protect freedom of expression
This bill entitled “Draft bill to amend the Danish Copyright Act” (Udkast til forslag til lov om ændring af lov om ophavsret), tabled by the Danish Ministry of Culture, aims to more strictly regulate the misuse of artificial intelligence in the creation of falsified content. Subject to its adoption by Parliament, this legislation would enter into force on 31 March 2026. The initiative, supported by a large parliamentary majority, responds to a growing need for regulation in a world where the generation of false images is increasingly simple and fast.
📣 FIRSH expert opinion: Denmark is therefore considering adopting a pioneering law to legally protect physical appearance against deepfakes, from the unprecedented angle of copyright. By explicitly erecting these biometric attributes as legally protected expressions, the project therefore departs from the traditional foundations of copyright — historically limited to the protection of intellectual works — to integrate personal identity into the broader field of intellectual property. Our identity would therefore be transformed into a product of which we have the property: This legislation could accelerate the commercialization of personal identity, for the benefit of celebrities, especially post-mortem. For the rest, offences already exist in the field of privacy, defamation or digital usurpation and many others that should already be applied!
➡️ Disney and NBC Universal attack Midjourney: Hollywood giants declare war on generative AIs
On June 11, Disney and NBC Universal filed a lawsuit against Midjourney, accusing Midjourney of training its AI models on copyrighted works in their iconic franchises without authorization. They accuse Midjourney of massive copyright infringement, in particular for generating images that strongly resemble copyrighted works from their franchises (such as Star Wars or Minions), without permission.
⚖️ The main charges:
- Unlawful use of protected content to train Midjourney’s AI.
- Creation of generated images that mimic characters, settings, or visual styles owned by Disney and Universal.
- Direct responsibility of Midjourney, despite their defense that only users are responsible for the images generated.
💰 What Disney and Universal are demanding:
- $300 million in revenue that Midjourney would have generated from their intellectual properties.
- $150,000 per infringed work in damages.
- A ban on using or reproducing their works in the future.
🌐 Broader issues:
This lawsuit could become a major legal precedent in the field of AI and copyright. It raises fundamental questions:
- Can AI be trained on protected works?
- Who is responsible for AI-generated creatives?
- How can we protect artists and studios from creative automation?
Litigation related to the training of generative AI on protected content is indeed multiplying in the United States. While no judgment on the merits has yet been rendered, the federal courts are beginning to closely examine the scope of fair use and to address a central legal issue: whether or not there is a licensing market for training data.
This point has become a strategic element in the defense of AI developers who rely on the theory of fair use, an exception to American copyright. As a reminder, American judges must assess “fair use” according to four criteria,
The four criteria of “fair use” are elements used in U.S. law to determine whether the use of a copyrighted work can be made without the authorization of the rights holder. These criteria are as follows:
🎯 1. The purpose and character of the use: Is it for educational, critical, journalistic, research, or parody purposes ? Is the use commercial or non-profit? Has the work been transformed (adding comments, analysis, etc.) or simply copied?
📚 2. The nature of the copyrighted work: Is the work factual (such as a scientific article) or creative (such as a novel or song)? More factual works are more likely to be used fairly.
📏 3. The quantity and substance of the work used: How much of the original work is used? Is it a small portion or an essential part (the “heart” of the work)? Even a small portion can be problematic if it is deemed essential.
💸 4. The effect of the use on the potential market for the work: Does the use harm the commercial value of the original work? Can it replace the original work on the market?
These criteria are assessed together, on a case-by-case basis, by the courts. No single criterion is decisive.
🔵The crux of the debate: Can training data be licensed?
In this type of case, AI developers argue that in the absence of a structured market for the licensing of training data, no economic harm can be established.
🔵An increasingly contested position…
The defence based on the absence of a licensing market is weakening, as several rights holders, such as the New York Times or the Associated Press, have already entered into paid agreements governing the use of their content for AI training.
And that changes everything: if there exists, even partially, a business model based on the transfer of rights for training, then an unauthorised use could be considered to be detrimental to this potential market. This weakens the fair use defence.
🔵The Copyright Office’s opinion
The debate is all the more sensitive because the U.S. Copyright Office itself has been encouraging AI developers to explore licensing solutions when they are available for several months.
📣 FIRSH expert opinion: Faced with the multiplication of unauthorized uses of works protected by generative AI, the High Council for Literary and Artistic Property (CSPLA), in its latest report published on July 2, 2025, and the Senate in its report on creation and AI published on July 10, 2025, both propose the establishment of a real licensing market for the use of content for training purposes.
The aim is to ensure the protection of copyright and guarantee fair remuneration for creators, while providing a legal framework for access to data for AI players.
At FIRSH, we support authors, performers and rights holders in the negotiation, structuring and legal security of these licenses, at the intersection of copyright and technological innovation, but also developers of AI tools who have chosen to remunerate artists for more reliability, transparency and ethics. Trusted AI, which also secures the training of their AI tools, will allow the arrival of even more efficient and unbiased AI tools!
For further information👉 Disney and Universal Sue Midjourney: Navigating AI Copyright Challenges and Fair Use Implications | ArentFox Schiff
Link to the Assignment 👉: https://www.documentcloud.org/documents/25972151-disneynbcu-v-midjourney/
➡️ Article 4 of the AI Act: the European Commission has provided clarifications
The European Commission recently published an FAQ dedicated to Article 4 of the IA Regulation, specifying the requirements for competence in artificial intelligence.
🔵 The objective : to strengthen the AI culture in organizations, beyond simple awareness modules.
Since 2 February 2025, Article 4 requires developers, deployers, users and service providers of AI systems to control the systems they design or use, according to their respective responsibilities. This requirement is not limited to high-risk systems: it applies whenever a system may affect rights, security, or access to essential services.
🔵 Clarifications provided by the FAQ
- Mastering AI is based on skills, not just knowledge. Individuals must understand the capabilities, limitations, and risks of the systems they handle.
- Training must be differentiated according to roles. Developers, operational staff, legal managers or business lines: everyone needs a specific level of expertise.
- The obligations extend to third parties. Service providers and subcontractors must also be trained, not only internal teams.
- Roles, responsibilities and levels of risk exposure need to be mapped. The approach must be adapted, based on a concrete analysis of uses and issues.
🔵 AI governance: much more than a training program
The FAQ sends a strong signal that AI literacy is becoming a pillar of systems governance, along with transparency, security and ethics.
To achieve compliance, organizations must:
- Define internal governance rules governing the development and use of AI;
- Implement continuous improvement processes, to ensure that systems remain understandable, reliable and aligned with regulations;
- Document the actions taken, with clear evidence of training, monitoring and management of AI skills.
📣 FIRSH expert opinion: As the use of AI systems is becoming unavoidable, it is imperative to train in them. Without a structured approach, organizations expose themselves to legal, operational and reputational risks. FIRSH supports developers of AI solutions in their compliance with the AI ACT (even if many provisions have not yet come into force)
Learn more 👉 https://digital-strategy.ec.europa.eu/fr/faqs/ai-literacy-questions-answers
➡️ Cognitive debt and artificial intelligence: lessons from an MIT neurophysiological study
A recent MIT study looked at the cognitive impact of using artificial intelligence tools like ChatGPT as part of an intellectual exercise.
The aim is to measure how technological assistance influences brain activity and memory.
🔵Methodology : 54 students, divided into three groups (without help, with Google, with ChatGPT), took a writing test under electrodes in order to analyze their brain activity.
🔵Primary Finding: Participants who used ChatGPT showed reduced neural activity, especially in memory-related areas.
-
- 83% of them were unable to summarize their own work after the test.
- None of them was able to quote a single exact sentence from their copy.
The study highlights what the researchers call a “cognitive debt”, linked to excessive delegation to artificial intelligence. Like GPS addiction, which eventually weakens our spatial memory, using AI too early could harm our ability to memorize and think critically.
To find out more 👉 A connected world: podcast and replay show | France Culture
✨PERSONAL DATA✨
➡️ Access to professional emails: the Court of Cassation extends employees’ right of access
In a decision dated June 18, 2025, the Labor Chamber of the Court of Cassation affirmed that emails sent or received by an employee via his or her professional email account constitute personal data. As such, the employer is required to provide a full copy, including metadata (timestamp, recipients), unless the rights and freedoms of others are infringed.
This decision marks a turning point: it validates an extensive use of the right of access provided for in Article 15 of the GDPR, even when this right is exercised for evidentiary purposes, in particular after the termination of the employment contract.
🔵 Several voices are being raised to denounce a broader interpretation than that adopted by the CJEU in its judgment of 4 May 2023, aff. C-487/21, F.F. v. Österreichische Datenschutzbehörde, in particular with regard to the obligation to transmit emails in their entirety or the associated metadata.
🔵 A major challenge for companies: processing this type of request involves significant technical efforts (selection, anonymization, contextualization) and could pave the way for an increased judicialization of the right of access.
To read the full decision here 👉 , Court of Cassation, Social Chamber, June 18, 2025, No. 23-19.022
📣 FIRSH expert opinion: The firm masters the issues related to the protection of personal data. We support our clients in complying with the GDPR, securing the data collected, and managing incidents such as data breaches or leaks.
➡️Deletion of defamatory content by the host provider in the context of an accelerated procedure on the merits
On February 26, 2025, the Court of Cassation issued two important decisions on the removal of defamatory content online via the accelerated procedure on the merits. One (n° 23-16.762) concerned anonymous videos on YouTube, the other (n° 23-22.386) defamatory tweets already condemned.
➡️ The first case (n° 23-16.762): anonymous defamatory videos on YouTube
In this decision, the Court of Cassation recognizes that a judge can order a host to remove defamatory content published anonymously, even in the absence of an adversarial debate with the author, provided that the content is manifestly illegal.
The case pitted several Monegasque personalities against Google Ireland, after the publication on YouTube of anonymous videos seriously accusing them. In the absence of identification of the perpetrator despite a complaint with the constitution of a civil party, the victims went to court under the accelerated procedure on the merits (Article 6-I-8 of the LCEN) to request the deletion of the videos and the communication of identification data.
🔵 Court of Appeal’s response :
It dismissed the plaintiffs’ appeal, considering that, without adversarial proceedings, the judge could not rule on the unlawful nature of the remarks, and that the mere attack on honour was not sufficient to characterize manifest damage. It refused to examine the proportionality of the measure sought, on the ground that the applicants had not demonstrated that it was impossible to identify.
🔵 Answer of the Court of Cassation :
The Court overturned the appeal decision, recalling that a judge can perfectly well order the removal of manifestly illegal content when the author has voluntarily organized his anonymity (via fictitious data and VPN). It emphasises that the seriousness of the damage to reputation requires an effective balancing exercise with freedom of expression.
It criticises the trial judges for not having investigated whether the removal was not proportionate, given the fact that the author was unreachable, and that the host (Google) did not have reliable identification data, in breach of its legal obligations (Article 6-II of the LCEN).
🔍 Critical focus: the requirement of “manifestly unlawful damage” in question
The Court made the removal conditional on the manifestly unlawful nature of the content, rejecting the idea that a simple attack on honour was sufficient. This reading imports the criteria of the 1881 press law (proof of truth, good faith, etc.) in a civil procedure that is supposed to be autonomous. It weakens the recourse of victims in cases of anonymity, even though the LCEN aims to act urgently against the host, without the need to criminalize the remarks.
📣 FIRSH expert opinion: The judgment confirms that a host can be forced to remove defamatory content even without adversarial proceedings, if the illegality is manifest and if the author is hiding. It imposes a rigorous control of proportionality on trial judges, to prevent anonymity from creating digital impunity. A decision that paves the way for better judicial protection… while leaving a legal uncertainty on the notion of “damage” within the meaning of the LCEN.
➡️ The second case (n° 23-22.386): reiteration of defamatory tweets already condemned
A recent decision of the Court of Cassation recalls that digital platforms have the obligation to remove defamatory content already judged as such, as soon as they are seized of it. The case concerned a Franco-Moroccan boxer reiterating on X (formerly Twitter) defamatory remarks against a senior official of the Kingdom of Marco that had already been convicted by the courts.
Considering that defamatory remarks already condemned were being disseminated again, the victim requested their removal from Twitter on the basis of Article 6-I-5 of the LCEN law. In the absence of a satisfactory response from the platform, it took legal action to obtain their deletion, prohibit any identical republication, identify the author of the account and obtain compensation.
🔵Court of Appeal’s response :
The Court of Appeal dismissed the plaintiff’s claim, on the grounds that the request was not directly aimed at the author of the comments, but at the hosting provider (here, X, formerly Twitter). According to the appeal judges, no adversarial hearing was possible, since the author’s direct responsibility was not at issue in these proceedings. In the absence of such a debate, the court held that it was not possible to assess the reality of the infringement.
🔵Answer of the Court of Cassation:
The Court of Cassation, seized by the plaintiff, overturned the appeal decision and clarified the scope of the obligations imposed on the hosting providers. It is based on Articles 6-I-7 and 6-I-8 of the LCEN, as amended by the law of 24 August 2021, and on the interpretation of the European “e-commerce” directive by the Court of Justice of the European Union.
It recalls that a host is not subject to a general obligation to monitor content, but that a court can perfectly well order it to delete or block access to a publication whose content is identical to that of a message already declared illegal (in this case, comments deemed defamatory in a previous decision).
This blocking does not constitute excessive censorship, provided that:
- the content in question is strictly identical to that already deemed illegal,
- the host does not have to make an independent assessment of the defamatory nature of the comments,
- The request for withdrawal is based on a prior judicial decision.
In this case, the Court of Appeal should therefore have verified whether the comments disseminated on the Twitter profile were indeed identical to those for which the author had already been convicted, which it did not do.
📣 FIRSH expert opinion: This decision of the Court of Cassation clarifies a sensitive point: platforms like X are not required to judge for themselves what is or is not defamatory, but they may be forced to remove content already recognized as illegal by a court decision whether or not the author of the defamatory statement is harmed at the instance. FIRSH returns to this subject and its practical implications in more detail in its white paper.
To read the full decision here 👉 , Cass. 1st civ., February 26, 2025, RG n°23-22.386
✨AI & INTELLECTUAL PROPERTY✨
➡️ PUBLICATION OF THE REPORT ON THE REMUNERATION OF CULTURAL CONTENT BY ARTIFICIAL INTELLIGENCE
Report on remuneration for cultural content by artificial intelligence is published
The High Council for Literary and Artistic Property (CSPLA) has published the report of its mission on the remuneration of cultural content exploited by artificial intelligence systems. Led by Alexandra Bensamoun and Joëlle Farchy, this report proposes a cross-disciplinary legal and economic approach to regulate the use of works in the training of AIs.
🔵 A clear starting point : AI systems feed on vast volumes of cultural data, often protected by copyright. While creators can theoretically oppose it via an opt-out mechanism, this is currently insufficient to ensure real protection and fair remuneration.
🔵 The report calls for more transparency on the sources used, in order to guarantee the effectiveness of the right to object and to better distribute the value generated.
- On the legal level, several levers are proposed:
- the establishment of an AI mediator to facilitate negotiation between developers and rights holders,
- tools to strengthen the litigation framework, such as the presumption of use, class actions or the injunction for disclosure of evidence,
- and a contractual toolbox for equitable rights management.
- From an economic point of view, the report recommends the creation of a licensing market, based on negotiation, with proportional or flat-rate remuneration depending on the usage. This structuring could rely on technical intermediaries and a content organization driven by rights holders, to meet the requirements of the AI ecosystem.
🔵 Finally, the report insists on the need to fully involve creators in the value chain, recalling that AI models themselves depend on the richness and diversity of protected content to maintain their relevance.
📣 FIRSH expert opinion: The CSPLA report marks an important step in the regulation of the use of AI in the face of copyright. It offers concrete tools to structure a real licensing market, reconciling legal certainty for developers and fair remuneration for creators. At FIRSH, we support creative and tech players in the implementation of contractual strategies, traceability mechanisms and business models adapted to the rapidly changing AI ecosystem.
➡️ Senate report: “creation and artificial intelligence”
⚠️On July 10, 2025, the Senate’s Committee on Culture, Education and Communication published a report entitled “Creation and artificial intelligence: from predation to value sharing”, resulting from a fact-finding mission led by Laurent Lafon, with Agnès Evren, Laure Darcos and Pierre Ouzoulias.
In the context of this work, the rapporteurs make an unequivocal observation: “Generative artificial intelligence models, largely developed by non-European actors, have been fed by a considerable mass of cultural content, texts, images, sounds, videos without prior authorisation or remuneration from the rights holders.”
🔍 Faced with this situation, described as “free and systemic capture of cultural value”, the rapporteurs envisage the introduction into French positive law of a presumption of data use: in the presence of a set of indicators, it would be up to AI providers to demonstrate that their models have not been trained from protected works. Inspired by the work of Professor Alexandra Bensamoun, this reversal of the burden of proof would constitute a structuring legal instrument to restore the balance between creators and technologists.
🔍 The report also envisages the creation of a tax based on the turnover of AI players, designed as a way to rebalance economic relations and encourage the conclusion of licensing agreements between rightholders and model developers. The report also denounces what it describes as “unsustainable free data”, the result of a lack of regulation combined with an underestimation of the stakes by the cultural sectors themselves. He warns against the head-on competition between human works and generated productions, which threatens certain creative professions in the long term.
🔍 Finally, the senators call for an ambitious European response: to build a model for the governance of cultural data, guaranteeing respect for rights, technological sovereignty and the sustainability of creation.
Without a quick reaction from Europe, the senators warn that Europe risks remaining the “useful idiot” of a digital order imposed from outside.
To find out more 👉: Creation and AI: from predation to value sharing – Senate
✨INTELLECTUAL PROPERTY ✨
➡️Trademark law: Figurative trademark representing a checkered pattern and designating, inter alia, clothing – Lack of distinctive character
A German company, owner of figurative trademarks representing an “argyle” pattern (black tile crossed by white lines), sued two foreign companies marketing similar socks for infringement before the Paris Judicial Court. In response, the defendants sought the cancellation of those marks on the ground of lack of distinctive character.
The judges agreed with the defendants. The motif in question, although stylized, is perceived as a simple decoration common in the fashion industry, and not as an indicator of commercial origin. The relevant public — average consumers — will see it as nothing more than a banal decorative pattern, even if it is represented by only one tile.
Nor has the applicant succeeded in proving the acquisition of distinctiveness through use. The evidence provided (sales, polls, presence on social networks) was deemed insufficient, particularly for the European Union, where the evidence must cover the entire territory.
🔵 Accordingly:
- The French and European designations of figurative trademarks are cancelled for lack of distinctiveness in classes 18 and 25 (leather, clothing).
- The revocation of rights is pronounced for classes 3 and 14 (cosmetics, jewellery), for lack of genuine use.
- The semi-figurative trademark “THE SIGN OF STYLE” is also down.
- Finally, the infringement claim is dismissed: although the signs at issue share some visual and conceptual similarity, the addition of the verbal elements “Burlington” or “Burlington The Original” in the claimant’s marks is sufficient to rule out any likelihood of confusion.
📣 FIRSH expert opinion: This decision is a reminder that classic decorative motifs in the fashion sector can only be monopolized as trademarks if they have acquired a real distinctive character in the eyes of the public.
To read the full decision here 👉 TJ Paris 19 March 2025 19/11680
➡️Trademark law: Refusal of a pattern mark that does not significantly depart from the norms and customs of the leather goods sector
The application for an extension in France of an international trademark consisting of a three-dimensional repetitive pattern — composed of triangular-shaped relief elements with rounded edges, arranged in a staggered pattern and reminiscent of drops of water or corn kernels — was rejected for lack of distinctiveness. The sign referred to goods in the luggage and leather goods sector.
In accordance with the settled case law of the CJEU, a three-dimensional trademark can only be registered if its shape departs significantly from the usages of the sector concerned, in order to be perceived by the public as an indicator of commercial origin, and not as a mere ornament.
This principle also applies:
🔵 figurative trademarks representing a product in two dimensions,
🔵 and surface patterns affixed to products, as in this case.
However, the claimed pattern does not differ significantly from the shapes commonly used in the leather goods sector for purely decorative purposes. It is a banal variant of textures already present on the market, without a sufficiently distinctive element to attract the consumer’s attention as a sign of identification of an origin.
As a result, the distinctive character is lacking, and the trademark therefore does not appear to be such as to identify the commercial origin of the goods offered. Protection in France was therefore legitimately refused.
To read the full decision here 👉 Paris Court of Appeal, pole 5, 1st ch., March 26, 2025, 23/13309
➡️Design law: Originality of jewellery with a snake’s head paved with precious stones – Infringement rejected, but acts of parasitism blamed
An Italian jewelry house claimed copyright on several pieces of jewelry from its iconic “Serpenti seduttory” collection, featuring a stylized snake’s head with eyes encrusted with precious stones.
🔵 It had sued a competitor company selling similar jewellery representing snakes, both online and in shops, for infringement.
🔵 In the alternative, it also invoked acts of parasitism.
The court recognized that the jewels invoked presented sufficient originality, based on precise aesthetic choices reflecting the imprint of the personality of their creator.
However, the infringement claim was rejected. The disputed jewellery marketed by the defendant did not contain either the characteristic elements of the protected jewellery or its specific combination.
On the other hand, the application on the ground of parasitism was granted. The snake has been a strong sign of identity for the Italian house for decades, highlighted by sustained communication, exhibitions and significant marketing investments. By systematically drawing inspiration from this symbolism in a series of pieces of jewellery without copying them identically, the defendant sought to take advantage of the reputation and individualised economic value of the ‘Serpenti sedutto’ collection, without assuming the equivalent investment efforts.
It thus placed itself in the commercial wake of the plaintiff, characterizing a faulty parasitic behavior, even if the snake theme remains, in itself, in free use in the sect
To read the full decision here 👉 Paris Court of Appeal, pole 5, 2nd ch., 31 January 2025, 23/05552 and 23/07023
📣 FIRSH expert opinion: The firm regularly advises its clients on their brand strategy. We advise and defend our clients to protect their trademarks, whether at the time of registration, in the event of a conflict, or to assert their rights in the face of infringement or dispute.
✨FIRSH ECOSYSTEM✨
➡️ DARE AI Plan
Launch of the national plan “Dare AI”:
This is an unprecedented initiative to accelerate the adoption of artificial intelligence in all companies, regardless of their size or sector.
Supported by the government and a network of AI ambassadors across the country, this plan aims to make AI tangible, accessible, and useful for SMEs, micro-businesses, and mid-sized companies.
The objective: to move from awareness to action, and to make AI a lever for growth for all.
✨ “FIRSH” NEWS✨
➡️Firsh took part in the France Transports Ambition Day: AI and data at the heart of tomorrow’s mobility
Last Monday, our partner Claire Poirson took part in the Ambition France Transports day at the Ministry of Transport, invited by Ministers Philippe Tabarot, Minister for Transport, and Clara Chappaz, Minister Delegate to the Minister of the Economy, Finance and Industrial and Digital Sovereignty, in charge of Artificial Intelligence and Digital Technology, dedicated to innovation in mobility.
Organised around round tables and collaborative workshops, the day brought together public and private experts to imagine transport in 2040. Claire Poirson spoke in the workshop dedicated to the use of AI and data, alongside major players in the sector.
🎯 The common ambition : accessible, supervised and ethical data, to feed frugal and responsible AIs for the benefit of users.
📣 FIRSH expert opinion : regulating is not the same as slowing down. It is to lay the foundations for a lasting trust between technology and society. At FIRSH, we support this transition with ethical, innovative and committed legal expertise.
➡️Firsh has assisted clients on the following projects, among others:
🔵In the context of the protection of a trademark for a “trademark” summary procedure on the basis of L716-4-6 of the Intellectual Property Code.
🔵 As part of an advisory mission to a company on the implementation in France of Directive (EU) 2019/882 of 17 April 2019, known as the “European Accessibility Act”, on accessibility requirements applicable to products and services.
📢 To follow us on LinkedIn and receive our newsletter, click here: https://www.linkedin.com/company/firshlaw/ .
📢 There is no direct collection of your personal data and therefore no emailing by Firsh!