The Legal Affairs Committee of the European Parliament (JURI) has published a report entitled “Copyright and generative artificial intelligence – opportunities and challenges”. At first glance, this may seem routine – papers of this kind emerge in Brussels almost on a weekly basis. This report is different, though. It addresses an issue that has preoccupied legal scholarship for decades and has never really been resolved: The tension between a copyright system rooted in the age of the printing press and a digital technology that operates with types of works which simply do not fit into the traditional understanding of copyright. In this respect, it is no exaggeration to speak of a turning point in copyright law.
What is it about?
At the heart of the matter is a conflict that is as crucial for the competitiveness of European businesses as it is for the future of the media and creative industries. On the one hand, Europe wants to catch up in artificial intelligence (AI) innovation and narrow the gap with the US and China. That is not possible without powerful models and large volumes of high quality training data. On the other hand, media organizations, publishers, the music and film industries, platforms, and individual creators are providing these training data on a massive scale – often without their consent and without remuneration.
Generative AI has already become an integral part of products and processes in many companies – from the creation of texts, images and software code to entire news sites. Behind this technological performance lies a simple but legally sensitive question: What were these systems trained on? The answer is usually on content created by journalists, authors, photographers, musicians, and other creatives – and therefore, as a rule, content protected by copyright. For good reason: For creatives, copyright is often the only legal basis on which they can control the use of their works and monetize their creative effort.
Historically, European copyright law is designed to protect individual works of authorship against unauthorized reproduction and distribution – that is, against traditional acts of copying. The debate on intellectual property protection for computer programs already showed that this logic only fits modern digital technology to a limited extent. Software that is constantly modified, combined, and executed in complex systems could only be fitted into the existing categories with considerable doctrinal effort; unsurprisingly, it still has its own section with special rules in the statute.
AI – particularly the training of generative AI – fits this proprietary concept even less. AI systems depend on processing not only “data” but specifically copyright-protected works for training purposes in order to be powerful and to deliver high quality results. Against this copyright backdrop, a number of key legal questions arise: May an AI system, without consent, scrape large quantities of protected works from the internet, analyze them, and generate new content from them? And if so, under what conditions, with which transparency obligations, and with what remuneration for the creative sector? First courts – for instance Munich Regional Court at the end of last year – have already held that certain forms of AI training with protected works and the use of the outputs are unlawful. European policymakers are now following suit.
The core message
AI systems do not work with a few isolated works, but with billions of digital copies. They analyze patterns, abstract content, and generate new outputs from them – in ways that can no longer be neatly mapped onto the classic copyright categories of reproduction, distribution, or adaptation.
Against this background, the Legal Affairs Committee reaches a strikingly clear conclusion: Existing copyright law is no longer adequate, in the age of AI, to protect all relevant interests. It is no longer sufficient as the sole regulatory framework. The report puts forward the idea of a standalone, complementary legal framework for AI and AI training.
At the heart of this proposed framework is the idea of a functioning licensing infrastructure: Right holders should be able to license their content specifically for training purposes; AI providers, in return, should be able to operate under clear and reliable rules. The aim is to strengthen the bargaining position of media companies and creatives, while enabling legally-secure use for technology companies – an attempt to relieve the overburdened traditional copyright system by supplementing it with specific AI rules.
This is notable because it entails a long overdue admission: The decades long strategy of “reading” ever new digital phenomena into old legal provisions has reached its limits.
The details
Opt out for right holders
A key element of the new framework is a standardized opt out system. In the future, right holders will be able to mark their content with machine readable signals that unambiguously mean that this content may not be used for training generative AI. The report envisages standardized technical formats for this and a central register, ideally hosted by the European Union Intellectual Property Office (EUIPO). Opt out declarations could be collected there and queried by AI providers. For media organizations, publishers, and major platforms in particular, this would, for the first time, create a technical and legal instrument that allows them actively to control whether and to what extent they feed their content into the AI data stream – instead of being mere bystanders to the training process.
Far reaching transparency obligations
The proposed transparency obligations for AI providers are particularly far reaching. In the future, they would be required to disclose which copyright protected content they have used to train their models and to report this information to the EUIPO. The office would receive the data, inform right holders, and monitor compliance with these obligations.
If a provider breaches these transparency requirements, a rebuttable presumption would apply. It would assume that all relevant works in a given field have been used for training – with the result that claims for injunctive relief and damages can be enforced considerably more easily. For companies that develop and provide AI models, this amounts to a paradigm shift: Training data will no longer be merely a technical and competitive factor, but a highly regulated asset, the use of which must be comprehensively documented and explained to authorities and right holders.
Expanded territoriality principle
On top of this comes a deliberately broad territorial scope. The report proposes developing the traditional principle of territoriality so that EU copyright law applies whenever an AI system is offered or used in the EU internal market – irrespective of where the training took place. A provider from the US or Asia could therefore not simply argue that its models were trained outside Europe. The decisive factor would be access to the European market. The goal is a “level playing field”: European and non European providers will be subject to the same copyright requirements if they offer their AI services in the EU.
Remuneration for authors
The report makes it clear that media companies and creatives should share in the economic success of generative AI. Various compensation mechanisms are under discussion, including retroactive remuneration for past training uses during a period when there was no developed licensing market. JURI rejects a global flat rate model; instead, remuneration will be proportionate and negotiated on the basis of relevant factors.
As an interim option, the report even mentions a flat levy of 5 to 7% of AI providers’ global turnover for the use of European content. Whether such a rate is politically achievable remains open. The underlying message is, however, clear: Significant redistributive effects are expected, and there is serious consideration of a dedicated remuneration regime for generative AI – another indication that traditional copyright law, on its own, is no longer sufficient as the steering instrument.
Protectability of AI outputs
Another central issue is the legal status of AI outputs. The report reaffirms the principle that, under European doctrine, copyright requires a human intellectual creation. Content generated purely by AI does not meet this requirement and should therefore, as a rule, be treated as being in the public domain and labeled accordingly.
Recent case law supports this approach: Munich Local Court has held that AI generated logos do not enjoy copyright protection. Neither mere prompting nor the selection between several AI suggestions is sufficient as a human creative contribution. For businesses, this is ambivalent. On the one hand, content generated purely by AI can hardly be protected on an exclusive basis, which has implications for brand building and content strategies. On the other hand, treating such content as being in the public domain creates a large pool of freely usable material – opening up new opportunities for business models, but also new competitive dynamics. In practice, it will become increasingly important to document the human contribution to creatively shaped outputs in order to substantiate protectability.
What does this mean – and for whom is it relevant?
Legally speaking, the JURI report is “only” a draft parliamentary resolution. Once adopted in the European Parliament, it will become a resolution which – unlike, for example, the AI Act – does not have immediate legal effect. It is a political call on the European Commission to clarify existing copyright rules and, where appropriate, to propose new legal instruments. The right of legislative initiative lies with the Commission; through resolutions of this kind, Parliament sets the agenda.
Nevertheless, such reports are a reliable indicator of the direction in which future legislation is likely to develop. Above all, the underlying diagnosis is clear: Copyright law as we know it is no longer sufficient in the age of AI to strike a balanced relationship between innovation, competition, and the protection of creative output.
For businesses, the report is therefore less an academic exercise than an early glimpse of the conditions under which AI will, in the future, operate in Europe. AI providers and technology companies will have to reconsider their data and documentation strategies. Those training models today will, in the future, have to demonstrate on which data the training was based, which rights attach to those data, and how opt outs have been respected. At the same time, a licensing and remuneration strategy will be needed that takes account of costs, risks, and opportunities – for example through sector specific or collective licenses. Business models, pricing structures, and market entry strategies for the EU will need to adapt accordingly.
Media companies, publishers, platforms, and other right holders receive a clear signal from the report that their interests are not only being heard at EU level but will be translated into concrete regulation. They should decide at an early stage whether they wish to withdraw their content from AI training entirely or in part, or conversely, whether they intend actively to feed it into emerging licensing markets. In parallel, they will have to consider how to position themselves institutionally – via collecting societies, trade associations, or proprietary rights platforms – in order not to remain mere spectators in future negotiations.
Finally, the report also affects all companies that “only” use AI. Those deploying generative AI in products, marketing, software development, or internal processes will need to look more closely in future: How does the relevant provider handle training data? What assurances are given regarding the lawfulness of training and the observance of opt outs? How are liability risks apportioned? AI governance will thus become a cross cutting task in which copyright issues sit alongside data protection and the AI Act as a fixed component.
What should companies do now – and in any event already?
- Generative AI should only be used where it can be ensured that the output does not infringe third party rights. In practice, this means, for example in relation to AI generated code, establishing systematic licensing, and compliance scans to identify covertly copied open source components or proprietary code. The same applies to texts, images, music, and designs: Companies should put in place contractual and technical safeguards to ensure that the AI services they use either rely on lawfully licensed training data, or provide appropriate indemnities and guarantees – and that these are not just “on paper”, but are underpinned by processes and audits.
- It is advisable to establish a bespoke internal AI usage framework. This includes an internal policy specifying which systems may be used for which purposes, which content must not be entered into external AI services (for example, confidential or highly sensitive data), and which review steps must be completed before AI generated content is published externally. Responsibilities – for example within the legal department, IT, and business units – should be clearly allocated.
- AI providers and major users should start documenting their data and training strategies now. Those who train their own models should record which data sources are used, what rights attach to them, and how they respond to any opt outs. This documentation will in future become important not only to regulators but also in dealings with business partners and investors.
- Companies that produce substantial proprietary content – media organizations, platforms, brand driven businesses – should decide in the near term whether, and on what terms, they want to make their content available for AI training. This includes examining technical opt out options, but also developing scenarios for how content could be fed into licensing or remuneration models as and when corresponding markets emerge.
- Existing contracts with AI providers, software vendors, and service providers should be reviewed, and new contracts should include clear provisions on copyright, training data, liability, and indemnities. The more legislators demand transparency and documentation, the more important enforceable contractual assurances will become.
- Finally, it is sensible to engage actively with the regulatory process – whether via industry associations, business networks, or consultation procedures. Those who provide feedback early can help to ensure that future rules are workable in practice and do not lose sight of the realities of business.
