Landscape Mapping: Law Cases & Initiatives

On 10 March 2026, the European Parliament Plenary adopted by a large majority a set of proposals on the Report of the Committee on Legal Affairs (JURI) on GenAI and copyright.

This is the first European policy text on the collective management of AI in Europe, calling on the European Commission to facilitate voluntary sectoral collective licensing agreements.

The main demands of the European Parliament are summarized as follows:

• Mandatory transparency: AI providers must keep detailed records and publish detailed lists of the protected works they use to train LLMs.
• Fair remuneration: Creators must be fairly and proportionately remunerated for the use of their data from productive AI models, including the use of works from previous years.
• Right to opt-out: Copyright holders must retain control and be able to exclude their works from data mining for training AI models.
• Human creation: Content generated entirely by Artificial Intelligence should not be protected by copyright.
• Rebuttable presumption of infringement: In the event that an AI provider fails to comply with transparency obligations, a presumption should be triggered that it has infringed intellectual property rights, with the burden of proof being on the AI ​​provider.

Czech music collecting societies OSA and Intergram introduce new fees for AI usage

New royalties for the use of Artificial Intelligence – the major Czech collecting societies (OSA and Intergram) have amended their royalty policy, effective from 2026, introducing new payments for the use of copyrighted works in connection with AI. These regulations concern both the reproduction of works for the purposes of machine learning and content generation (output generation). The royalty rate was set at a high level – at 9% of the annual turnover of the service that performs machine learning and/or generates content through GenAI.

The debate in Denmark on the use of copyrighted material for training generative AI models is evolving, shifting its focus from whether compensation is due to creators, to determining the structure and details of a remuneration system.

The debate seems to focus on a number of remuneration systems, three of which deserve particular attention:

First, an opt-out mechanism is proposed as a way of licensing (or obtaining consent) for the use of protected works for training in AI.

The second proposal is to remove the provision for commercial text and data mining (TDM) in favor of introducing a legal license for the use of a work for training AI models.

A third option is to subject AI-generated content to a type of royalties paid to the community of creators.

Pricing (Fees / Distribution of Rights)

In particular, the group suggests that commercial exploitation and data mining (TDM) should not be done without permission and payment.

AI providers should pay fair remuneration to copyright holders.

The management of fees and distribution of these revenues is proposed to be done through expanded collective licensing systems, ensuring that revenues reach creators.

Arbitration Process

To resolve disputes between AI providers and ICOs:

It is proposed to give the Copyright Licensing Board (Ophavsretslicensnævnet) the power to resolve disputes related to licensing terms and the amount of remuneration.

Retroactivity of use

The discussion on retroactivity mainly concerns already “trained” material:

Under current EU law, the exception for text and data mining purposes is allowed, unless the creators have explicitly stated their reservation (opt-out).

For cases of mass use before the implementation of new regulations, there is a strong discussion (as documented in open letters from organizations in Denmark) about fair agreements covering the use of the material in EU markets.

Koda, the Danish CMO, initiated legal action against Suno, accusing it of illegally training its generative AI model on protected Danish music—such as hits by Aqua and MØ—without obtaining the creators' consent, without providing transparency, and without paying them royalties.

Koda's lawsuit is based on three core pillars:

• Consent:

AI developers must obtain explicit licenses before using human-created works to train GenAI models.

• Transparency:

Tech companies must disclose the source material used in their AI training data instead of relying on unverified "black box" algorithms.

• Compensation:

Creators must be compensated fairly for the use of their intellectual property, as generative AI tools now produce music that directly competes with human artists.

Koda's legal action highlights a severe financial threat: An economic report commissioned by the national industry estimates that unauthorized AI could cost the Danish music industry up to 28% of its annual revenues by 2030.

KODA's movement aligns with a broader global push by rights holders demanding stricter standards for AI services.

French Senate - “Darcos” Bill to reverse the burden of proof for the use of protected content in GenAI models.

Key details and guidelines:

• Presumption of Use: In any civil proceeding, a protected work is considered to have been used by the provider of a GenAI model, provided there is a “reasonable indication” of development, promotion or production of results.
• Rebuttable presumption: The AI ​​company can rebut it by proving that the specific work was never used for training, possibly resorting to a judicial expert opinion.
• Objective of the Law: The protection of intellectual property and the obligation of AI companies to have transparent records and a detailed record of the origin of their data.
• Progress of the Legislation: Representatives of intellectual creators and the music industry (more than 200 representative bodies ) continue to push for legislative completion in view of new discussions next October.

To enforce this model on the market, GEMA has taken significant legal action:

1.      Court victory against OpenAI: GEMA achieved a historic decision in the Munich Regional Court (GEMA v. OpenAI). The court ruled that storing lyrics in models (“memorization”) and reproducing them without permission constitutes a copyright infringement, obliging companies to seek official licenses.

2.      Legal action against Suno AI: On July 31, 2026, the Munich Regional Court (42nd Civil Chamber, Judge Elke Schwager) ruled that Suno did not have the rights to use music by songwriters and composers represented by GEMA. The Court found Suno liable for the unauthorized reproduction and communication to the public of the content and ordered it to cease using the works in question, to fully disclose its model's training data, and to pay the applicable financial compensation. 

3.      AI Charter: In parallel with the licensing model, GEMA published a charter of 10 principles, calling for complete transparency in training data and the protection of human creativity

The lawsuit filed by the CMO GEMA against OpenAI is currently at the stage of the first-instance ruling (in favor of GEMA), which is not yet final. OpenAI has appealed the decision.

In particular:

• Court & Date: The Munich Regional Court, in its decision dated November 11, 2025 (Case No. 42 O 14139/24), granted GEMA's claims for injunctive relief, information, and damages against two OpenAI group companies.
• Core Rationale: The Court held that the "memorisation" of song lyrics in OpenAI's language models constitutes reproduction, which falls under the right of reproduction pursuant to the German Copyright Act (Sec. 16 UrhG) and the InfoSoc Directive.
• Current Status: This decision is not yet legally final.

On 31/7/2026, the Munich Regional Court (42nd Civil Division, Judge Elke Schwager) ruled that Suno did not have the rights to use music from songwriters and composers represented by GEMA. Following the initial court ruling, Suno announced that it would appeal the ruling.

The court ordered Suno to disclose the revenues associated with this infringement and ruled that it would be liable for damages to GEMA. The amount has not yet been determined.

What makes this decision significant for the collective management sector:

• The court ruled that Suno copied and incorporated six songs represented by GEMA during model training, which constitutes an infringement under German law.
• Suno had argued that the actual training took place in the US and that its conduct was protected by fair use, with the US framework permitting unauthorized use for purposes such as research or satire, and separately challenged German jurisdiction altogether. The court’s willingness to conclude and reject this line of defense -rather than reject jurisdiction- is significant.
• Suno has announced that it will appeal, stating that it disagrees with the decision, which it says is based on a mischaracterization of how its technology works and how US law applies.

This is a first-instance decision concerning six specific compositions by GEMA member creators, and is not binding on US courts.

GEMA, the German CMO, had filed a lawsuit accusing Suno of illegally using protected works without permission to train the LMM model, while also alleging that the system can reproduce works that are very similar to the originals when given simple prompts.

Read more...

Proto Thema’s publications will supply content to Google News and the Gemini app, as well as providing a continuous stream of journalistic articles.

Proto Thema is part of Google’s AI pilot program, becoming the first Greek publishing organization to participate in this initiative. Proto Thema’s journalistic content, under the agreement, will be utilized in real time in Google News’ AI experiences and the Gemini application, offering users more accurate and up-to-date information.

The program aims to explore how AI can strengthen the connection with the audience and support high-quality journalism and focuses on testing experimental features designed to respond to readers’ habits.

Proto Thema is participating alongside a number of international publishing organizations - such as Der Spiegel, Financial Times, The Washington Post, The Guardian, El País, The Washington Examiner, Folha de S.Paulo, Infobae, Kompas, and The Times of India, among others - to explore how Artificial Intelligence can contribute to cultivating a highly interactive audience.

On 16/7/2026, the Plenary of the Parliament voted in favor of the draft law of the Ministry of Digital Governance and Artificial Intelligence, entitled "Measures for the implementation of Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (Artificial Intelligence Regulation) - Amendment of Law 4961/2022 (Government Gazette A’ 146) and other provisions", which establishes the National Framework for the Implementation of the European Regulation on Artificial Intelligence (AI Act).

With the completion of the parliamentary process, Greece is among the first member states of the European Union to establish a comprehensive national implementation mechanism for the “AI Act”, confirming the country’s strategic choice to combine innovation with a clear institutional framework, which strengthens citizens’ trust, supports research and creates a stable environment for the development of Artificial Intelligence applications.

The new framework defines the competent implementing and supervisory authorities, organizes the cooperation and complaint mechanisms, provides for effective sanctions for violations and introduces targeted measures to strengthen research and innovation. At the same time, it creates a stable and predictable environment for Public Bodies, businesses and investors, harmonized with European law and the requirements of the new digital era.

Of particular importance is the operation of the regulatory sandbox, through which businesses, start-ups and research institutions will be able to test and develop Artificial Intelligence applications in real conditions, with the support of the competent authorities. In this way, the development of innovative solutions is facilitated, without compromising the safety and protection of citizens.

Among other things, the bill for the implementation of the "AI Act" provides for:

• Definition of the Personal Data Protection Authority (PDPA) as the Central Market Supervisory Authority.

• Definition of the National Telecommunications and Post Commission (EETT) as the notifying authority and establishment of a Center for Coordination and Know-how for Artificial Intelligence within it.

• Definition of effective, proportionate and dissuasive administrative sanctions for violations, with the possibility of imposing sanctions on public bodies as well.

• Development of a unified system for submitting and managing complaints.

• Operation of regulatory sandboxes (AI Regulatory Sandboxes), with particular attention to small and medium-sized enterprises and startups.

• Creation of a Unified Registry of Artificial Intelligence Systems used by Public Administration.

• Establishment of a regulation, with criminal sanctions, prohibiting the removal of transparency labels from deepfakes content.

The *Like Company v. Google Ireland* case (Case C-250/25) is a landmark European legal dispute challenging whether generative AI infringes upon press-related intellectual property rights. It examines whether training AI on copyright-protected news content and displaying summaries in chatbot responses violates rights of reproduction and communication to the public.

Legal Dispute The Like Company v. Google case (Case C-250/25) is pending before the Court of Justice of the European Union (CJEU) and is at the stage of oral proceedings and consideration by the Advocate General, with the final decision expected in 2027

Important Stages and Developments:

·         Referral (April 2025): The Budapest Regional Court (Hungary) submitted preliminary questions to the CJEU.

·         Hearings (March 2026): The CJEU held the first hearing, where Google Gemini practices were examined in detail.

·         Current phase: The case is in the process of issuing Opinions by the Advocate General and then a conference of the CJEU judges for the final decision.

The main legal questions examined:

·         Copying for training (Training): Whether the use of protected articles for training the LLM model constitutes an act of reproduction and whether it is exempt from the rules on text and data mining (TDM).

·         Chatbot outputs (Output): Whether the display of content from journalistic pages within the AI ​​chatbot’s responses constitutes “presentation to the public” and infringes the rights of press publishers under Directive (EU) 2019/790.

On 14 July, the Irish Parliament (Dáil Éireann) passed a bill to protect the copyright of Irish artists in the age of Artificial Intelligence, following a motion by politician and author Aengus Ó Snodaigh.

Aengus Ó Snodaigh has been a Sinn Féin Member of Parliament (TD) for Dublin South-Central since 2002.

The bill calls for transparency, mandatory consent and fair remuneration for creators to stop artificial intelligence systems from using human-made art and music for education without permission. A rare cross-party consensus was reached in the House, with the central message being that copyright should support technological innovation, not replace human work.The Ivors Academy, IMRO and Screen Composers Guild of Ireland (SCGI)nhave welcomed the passing by Dáil Éireann of Sinn Féin’s motion, Protecting the Copyright of Irish Artists in the era of Artificial Intelligence, describing it as an important step towards ensuring that music creators’ works cannot be used by generative AI systems without consent, transparency, licensing and fair remuneration.

• Main Demands on Government
• EU Presidency Action: Urges Ireland to use its position and influence during its Presidency of the Council of the EU to spearhead regulatory protections for creators' rights.
  Funding Safeguards: Ensures state and public arts funding explicitly supports human-created works rather than AI-generated output.

• On 10 June 2026, the Italian government gave preliminary approval to two implementing decrees under its AI Law (Law 132/2025), aligning national rules with the EU AI Act
• The framework confirms that AI may be used to support workplace processes, including organising and analysing work-related information.
• However, decisions affecting employees’ rights must always involve human judgment, and cannot be fully automated.
• In particular, hiring, dismissal, and disciplinary decisions must not be taken solely by AI systems, and a responsible individual must remain accountable.
• The decrees are not yet final and remain subject to parliamentary review before coming into force.

The AGCOM v. Meta Platforms case (Case C-797/23) is a landmark case for the press and copyright worldwide.

The Court of Justice of the European Union (CJEU) has definitively ruled in favor of the Italian telecommunications regulator, confirming that EU member states have the right to oblige major platforms (such as Facebook and Instagram) to negotiate and pay fair remuneration to publishers for the use of their articles.

The National Library of Norway and the collective management organization Kopinor are collaborating to make copyrighted books, newspapers and other materials legally available online, as well as for training Artificial Intelligence systems. This collaboration allows ordinary users with Norwegian IP addresses to read thousands of texts for free, while ensuring fair remuneration for the original authors and publishers.

The government was led to this decision following strong reactions and mobilizations by the country's creators and artists' associations.

The withdrawal and the next step in the institutional dialogue are characterized by the following:

·         Reason for Withdrawal: Intellectual property organizations complained that the proposed licensing system would legalize massive and illegal uses of works by AI companies. Despite the opt-out option provided for in the law, the cultural sector considered the regulation insufficient.

·         The Starting Point of the Law: It was the first attempt in Europe to implement a national framework for collective management of copyright specifically for the development of general-purpose AI. The aim was to circumvent the impossibility of concluding voluntary individual agreements with millions of creators.

·         Current Path (2026): Instead of collective licenses, the Ministry of Culture has initiated dialogue processes with the industry to find sustainable solutions. At the same time, at the national level, Spain is promoting strict rules against deepfakes and requiring the labeling of AI content with huge fines for violators.

Its goal is to enable AI companies to legally train their models, while ensuring transparency and remuneration for creators.
 

Key features of the model include:

• Covering All Stages: The license does not only concern the initial stages of training an AI model but also compensates creators for the use of the final produced works (output).
• Attribution Technology: The model is based on independent data tracking platforms such as Sureel. Thus, AI-produced music can be traced and attributed to the original human works that influenced it, allowing copyright to be paid.
• Combined Licensing: Although STIM usually only represents the copyright of the musical composition, the pilot also covers the master rights for a selected catalog.
• Pilot Phase (Opt-in): Initially implemented with selective agreements, such as the one with the Songfox platform. It covers works by creators who have given their express consent, with the aim of expanding the framework more widely in the future.
• Fee Structure: AI companies pay a combination of upfront fees and percentages of revenue generated from operating their service or distributing the content they produce.

Created by the Copyright Licensing Agency (CLA), in collaboration with the Publishers' Licensing Services (PLS) and the Authors' Licensing and Collecting Society (ALCS).

Key Features and Goals:

• Fair Compensation: Ensures that authors and publishers are fairly compensated when their works are used by AI systems, providing a solution primarily for creators who do not easily negotiate individual agreements.
• Legal Assurance: Offers AI developers a cost-effective and transparent way of licensing, protecting them from copyright infringement.
• Opt-in Model: Operates on an opt-in basis, meaning that publishers must explicitly choose whether to include their content in the program in order for it to be used.
• Context: This scheme operates alongside and complements potential direct agreements that publishers may enter into directly with AI companies.

This initiative is a response by the Organisations to the UK’s controversial – and subsequently withdrawn – government proposals for exemptions that would allow AI models to be trained without the permission or remuneration of the creators.

Damian Hinds MP has introduced the Automated Online Software (Access and Transparency) Bill, a Private Member’s Bill backed by the News Media Association, to crack down on the growing threat to digital businesses – including news publishers – from an overwhelming surge in anonymous bots that extract content, data and value from their websites.

The Bill will allow website owners to identify who is trying to access their website and why. This will give them the information they need to deter harmful or costly traffic, allow it if beneficial, or even point bot operators to opportunities to pay a fair price for the value of the content.

Bots are nothing new, but bot traffic has now overtaken human traffic for the first time – and 99 per cent of bot traffic is unwanted or unverifiable by the website owner.

Overall AI bot traffic rose by 300 per cent in 2025, with agentic AI traffic rising by almost 8,000 per cent. Media websites are among the primary targets, along with online retailers and cultural institutions.

News publishers have seen significant falls in web traffic as AI chatbots use bots to pull content from their websites in real time without permission, meaning the user never visits a publisher’s website.

The UK government has now officially abandoned its controversial plan to introduce a broad opt-out text and data mining (TDM) exemption for AI companies. The decision was made after huge backlash from the country’s creative industry, leading to a reversal of its original intention.

Current status (2026):

• Official finding: The government has published the relevant Report and Impact Assessment on Copyright and Artificial Intelligence, which “freezes” the changes, rejecting the opt-out model.
• Pending & “Wait-and-See”: No immediate legislative regulation is being promoted. The government chooses not to take a position until it assesses international developments, leaving room for the development of voluntary licensing agreements at the initiative of the industry itself.
• Protection against deepfakes: Instead of relaxing the framework, the government focuses on transparency for creators and considers separate protection measures for digital copies (deepfakes) and independent creators.