Archive

On 2 June, the European Commission holds its first workshop to identify text and data mining (TDM) opt-out protocols that GPAI model providers must comply with under the AI Act. The framing of the process is a negotiation between two groups: AI companies and rightsholders. That framing is wrong, and the consequences of getting this wrong will be felt far beyond either group.

This Is Not a Bilateral Trade Dispute

The protocols being discussed here will not just govern how a handful of large AI developers crawl a handful of major publishers. They will govern how automated systems interact with the entire public web.

That means: universities and open access repositories whose content underpins scientific research; libraries and cultural heritage institutions whose digital collections serve the public; individual creators, journalists, bloggers, and civil society organisations who publish online without the legal and technical resources of large commercial publishers; small businesses across every sector; and the open source communities whose work constitutes a substantial proportion of what is publicly available online.

None of these groups was meaningfully represented in the December 2025 consultation that forms the basis for the 2 June workshop. Of 153 respondents, 60% were rightsholders — dominated by book and press publishers — and just 7% were GPAI model providers. Everyone else was folded into an “other stakeholders” category that does not capture the full range of interests at stake.

A process structured as a negotiation between content industries and AI developers, aimed at reaching “general agreement” between those two groups, is not an adequate basis for decisions that will reshape access to and control over the public internet. The Commission is not setting commercial terms between two sectors. It is setting infrastructure-level rules for the web. That requires a broader mandate, broader participation, and broader scrutiny than this process has delivered so far.

The Commission’s Own Report Undermines Its Own Front-Runners

The summary report published ahead of the workshop records “a certain level of divergence across stakeholder groups, both in terms of technical assessment and overall positioning.” That is the headline finding. The three front-runners — TDMRep, C2PA TDM Assertions, and JPEG Trust Core Foundation V2 — are each described in terms that do not support endorsement.

GPAI model provider assessments of TDMRep are “predominantly critical or cautious,” citing limited adoption, uneven uptake across sectors, and scalability challenges. C2PA’s application to TDM opt-out signalling “remains experimental, with uncertain scalability and limited real-world deployment,” and the relevant functionality sits in third-party extensions rather than the core specification. JPEG Trust was not designed as an opt-out mechanism, is limited to JPEG files, and is acknowledged even by rightsholders to be at an early stage of development.

Measure 1.3(1)(b) requires protocols to be state-of-the-art, technically implementable, and widely adopted across cultural sectors. The consultation evidence does not show that any of the three front-runners meets all three criteria.

Privately Controlled Protocols Have No Place on a Regulatory List

The consultation report records clear opposition from AI developers and other respondents to privately controlled or proprietary protocols that create dependencies and distort competition. This concern is amplified when the protocols in question will affect the whole web.

Endorsing privately operated mechanisms gives those operators a state-sanctioned competitive advantage and forces compliance through commercial relationships rather than open standards. Open access publishers, research repositories, and smaller content producers who lack the resources to engage with proprietary systems on commercial terms will be structurally disadvantaged. Any protocol endorsed under this process must be open, community-governed, and free from control by a single private entity with a financial stake in its adoption.

Multiple Non-Interoperable Protocols Create the Problem They Claim to Solve

The report also records that multiple non-interoperable protocols operating simultaneously generate conflicting signals and legal uncertainty. If one protocol permits AI training at domain level and another forbids it at file level, there is no workable resolution. This problem scales with the diversity of the web. The broader the population implementing different combinations of mechanisms — and on the open web that population is enormous — the more frequent and consequential those conflicts become. Endorsing a list before interoperability frameworks are in place does not create certainty. It multiplies confusion.

What Should Happen on 2 June

The Commission should use the workshop to do three things. First, acknowledge openly that the scope of this process extends to the entire web ecosystem and commit to a second consultation phase that reaches the constituencies currently absent. Second, apply the criteria of Measure 1.3(1)(b) honestly to the current candidates and acknowledge that none of them yet qualifies. Third, give the IETF’s AIPREF working group — which is developing open, internationally recognised standards through a genuinely multi-stakeholder process — the space and time to deliver before any durable list is published.

The web belongs to everyone. The rules governing how it is crawled should be set by a process that reflects that.

On 28 January, the European Parliament’s JURI Committee will vote on MEP Axel Voss’s own-initiative report regarding ‘Copyright and Generative Artificial Intelligence’. While the report is framed as a response to the challenges of the AI era, it suffers from fundamental flaws that threaten to undermine European innovation and legal stability.

The Coalition for Creativity (C4C) joins other digital rights advocates, such as COMMUNIA, in calling on MEPs to reject the report. Here is why this proposal, in its current form, is a step backwards for the European digital single market.

The Fundamental Flaws in the Report

1. A Conflict with the AI Act

The report introduces transparency requirements that are technically impracticable and go far beyond the hard-fought compromise of the EU AI Act. It calls for “full transparency,” including itemised lists of every individual piece of copyright-protected content used for training and detailed records of all crawling activities.

Because large-scale models are trained on vast, heterogeneous datasets, generating such lists is practically impossible. These obligations would cripple European developers with compliance costs and technical complexity while foreign competitors operate without such burdens.

2. Creating “Ambiguity” Where None Exists

The report rests on the claim that there are “ambiguities” regarding how Text and Data Mining (TDM) exceptions apply to AI training. This is a move away from the original draft’s even more extreme position—that training fell entirely outside the copyright framework—but it remains problematic.

In reality, the European Commission has repeatedly confirmed that the existing TDM framework is the cornerstone for AI development in Europe. Reopening this settled law creates unnecessary market instability and harms researchers and developers of public-interest AI.

3. Unjustified Copyright Privileges for Media Content

One of the most concerning additions to the text is a call to grant press and news media rightsholders “full control” over the use of their content for AI training and downstream uses.

This would effectively remove press content from the scope of TDM exceptions by requiring “explicit consent” for use. There is no legal justification for elevating media content to a privileged status above all other copyrighted works, and there is a clear risk that such a move would threaten the preservation of our media memory for future generations.

4. Copyright Overreach

The report attempts to use copyright law as a “Swiss Army knife” to solve unrelated issues, such as deepfakes and privacy concerns. Copyright is a tool for protecting expression and rewarding creativity; it is not the appropriate regulatory mechanism for addressing personality rights or misinformation.

The Path Forward

The TDM exceptions are a vital part of the EU acquis that enables European innovation. By attempting to “gold-plate” the AI Act and fragment the copyright framework, this report undermines the EU’s goals for digital sovereignty and competitiveness.

This is why we have urged the JURI Committee to reject the report and ensure that European AI policy remains grounded in technical and legal reality.

Along with 83 organisations, the Coalition for Creativity (C4C) co-signed a joint statement on preserving net neutrality in the European Commission’s upcoming Digital Networks Act (DNA) proposal.

This is an important effort, supported by a broad-based set of stakeholders, to protect the Open Internet and take a firm stand against (concealed) network fees creeping into DNA.

Read our recent blog post, which outlines why the EC’s focus on creating a “level playing field” could lead to regulatory capture and a threat to the Open Internet.

The EC’s vague ideas of “cooperation” and “innovation” could lead to:

• Forced payments from content providers to telecoms
• Dilution of net neutrality through loopholes
• Regulatory bodies arbitrating financial flows, not defending users

C4C’s message: We must reject any “clarification” that enables pay-to-play models and reinforce net neutrality for all.

Make your voice heard by 11 July 2025! Contribute to the EC’s public consultation on the DNA.

Cross-posted from C4C’s LinkedIn page – see the original article

German MEP Axel Voss (EPP) shared his draft report on copyright and generative AI (2025/2058(INI)) last week. This own-initiative report of the EU Parliament JURI Committee, while lacking direct legislative force, carries some influence in shaping future EU policy on artificial intelligence and copyright law.  

The non-binding nature of this report should not diminish scrutiny of its proposals, as Parliament resolutions often inform subsequent Commission legislative initiatives and Member State positions.

This draft report is scheduled for discussion at the 15 July JURI Committee meeting (16:00-18:30 CEST). MEPs will have until 12 September (TBC) to table amendments to the report.

This analysis examines serious concerns with the report’s approach across technical feasibility, economic impact, legal consistency, and strategic alignment with EU competitiveness goals, arguing that the proposed framework contains fundamental flaws that would undermine European AI development and affect European consumers while failing to achieve meaningful protection for creators.

Technical Impossibility Meets Economic Reality

The Voss report’s core requirement for detailed itemisation of copyrighted works in AI training datasets represents a computational impossibility that would single-handedly eliminate European AI development. Modern frontier models like GPT-4 are trained on 13 trillion tokens from billions of sources, with the largest datasets containing 15+ trillion tokens. The computational overhead of tracking individual source attribution would increase training costs by 10-100x, making European AI development economically unviable. 

The mathematics are unforgiving: Training costs already reach $100 million for current models, with next-generation systems projected to cost $1-10 billion. Adding granular content tracking would require additional petabytes of storage and exponentially more computational resources. As Stanford Professors Mark A. Lemley and Bryan Casey explain in the Texas Law Review, “there is no plausible option simply to license all of the underlying photographs, videos, audio files, or texts” on which ML models are trained. If you were surprised at the cost of saying “Thank you” and “Please” to ChatGPT, as claimed by Sam Altman, just imagine the impact of the proposed obligation.

Legal scholars who have closely examined the AI Act’s transparency requirements consistently emphasise the impossibility of detailed content tracking. Dr. João Pedro Quintais of the University of Amsterdam’s Institute for Information Law (IViR) notes in his comprehensive analysis of the AI Act that “the goal cannot be for [General Purpose] AI providers to list all or most of the copyrighted material they are including in their training data sets in an itemized manner with clear identification of rights ownership claims, etc., as this would make this provision impossible to comply with”. Quintais further explains that “the low threshold of originality, the territorial fragmentation of copyright and its ownership, the absence of a registration requirement for works, and in general the poor state of rights ownership metadata demonstrate this impossibility.” 

The Voss report’s requirements would require tracking individual tokens across distributed training systems operating on thousands of GPUs, a technical impossibility that reveals the report’s fundamental misunderstanding of both AI architecture and existing legal frameworks.

Copyright Levies: Taxing Innovation to Fund Dysfunction

The numbers are damning: Past studies have shown that it costs 51.2 cents to collect each Euro for EU copyright levies, with only 53 cents of every euro reaching artists in Germany. This represents a catastrophic administrative failure that benefits bureaucrats more than creators. The system operates as what competition law scholars identify as “price-fixing cartels” with monopolistic positions.

Collecting societies exhibit profound structural inadequacies with “complete lack of accountability” and tariff-setting processes that can take up to 10 years. They “usually privilege top-selling artists and are not well suited to accommodate non-commercial music production,” creating massive coverage gaps. The distribution of levy revenues to recording artists amounts to less than €0.01 per album, a figure that exposes the system’s fundamental failure to serve its stated purpose.

The levy variations across EU countries range from €0.02 in Romania to €2.6 in France per capita, variations that cannot be explained by any underlying concept of economic harm. This arbitrary system taxes innovation while failing to compensate creators, representing the worst of both worlds for European competitiveness.

Scope Creep: Data Is Not Copyrighted

The Voss report contains fundamental conceptual errors that violate established legal principles distinguishing raw data from copyrightable expression across the globe. The EU Database Directive explicitly states that protection “does not extend to the data or material itself”, establishing a clear distinction between data compilation and the underlying data elements. This principle aligns with international jurisprudence, including the US Supreme Court’s Feist decision, which definitively established that “raw data does not satisfy the originality requirement” for copyright protection).

The legal foundation is clear: Copyright protects expressions, not ideas, procedures, or methods. The CJEU’s Infopaq decision requires that protected works be “original in the sense that it is the author’s own intellectual creation”. The Voss report’s conflation of factual data extraction with copyright infringement represents a worrisome expansion of copyright beyond its traditional boundaries.

The report’s approach creates confusing regulatory contradictions with the AI Act, which requires transparency about training data without presuming copyright infringement. The AI Act’s measured approach to copyright compliance directly conflicts with Voss’s presumption of infringement, creating legal uncertainty that violates due process principles and undermines the rule of law.

Fundamental Rights Versus Regulatory Capture: The Deeper Structural Failure

From a fundamental rights perspective, the Voss report violates multiple rights of the Charter of Fundamental Rights of the European Union: Article 11 (freedom of expression and information), Article 13 (freedom of arts and sciences), and Article 16 (freedom to conduct business).

AI training involves processing publicly available information to extract patterns and correlations. This represents a core aspect of freedom of information protected by the Universal Declaration of Human Rights and the European Convention on Human Rights. When the Voss report restricts AI research and development through licensing requirements and payment obligations, it fundamentally contradicts academic freedom and the established right to use available materials for legitimate research purposes and puts up another (artificial) barrier to the 5th Freedom, the free circulation of knowledge, at a time where the EU claims to promote it.

The economic dimension reveals an even more troubling pattern. MEP Voss’s approach creates what economists recognise as insurmountable transaction costs for licensing millions of individual works, while simultaneously generating windfall profits for rightsholders without corresponding benefits to consumers or broader social welfare. The mathematical reality of AI training datasets exposes the fundamental impossibility of meaningful individual creator compensation. As technology law expert Andrés Guadamuz observes, “for there to be a traditional licensing system, there has to be a value to each individual work that is worth the transaction costs, but if what you are going to get is 0.00000001% of an already low transaction cost, then most creators would not expect to see any money at all”. He aptly describes this as “the Spotify remuneration model, but a thousand times less money going to creators.” Basically, when training datasets contain billions of works, individual contributions become economically meaningless while administrative costs consume any potential distribution.

This represents a textbook case of regulatory capture, where legacy interests successfully manipulate the policy process to extract rents from innovation at the expense of European competitiveness and fundamental rights.

Conclusion: Defending European Digital Sovereignty

The Voss report represents regulatory capture by legacy interests seeking to extract rents from European innovation. Its technical requirements are impossible, its economic impact is destructive, its legal foundations are unsound, and its strategic implications are catastrophic for European competitiveness.

A proper policy response must reject the report’s fundamental premises, defend fundamental rights to information and innovation, and align copyright policy with European strategic autonomy goals. The EU’s choice is simple: lead the global AI revolution or become its casualty. The Voss report ensures the latter outcome while serving narrow interests at the expense of European innovation, competitiveness, and fundamental rights. The choice lies with European policymakers, but the consequences will impact European citizens, researchers, and businesses for decades to come.

Picture of Rapporteur MEP Axel Voss (EPP, Germany) © European Parliament – 4 June 2025 Generative AI and Copyright JURI Committee Workshop

Cross-posted from C4C’s LinkedIn page – see the original article

In Brussels, where policy language often wraps complexity in neutrality, few phrases are more deceptively comforting than “level playing field.” It evokes fairness, balance, an equal start. But when it appears in the European Commission’s Digital Networks Act (DNA) proposal, it risks becoming a euphemism for regulatory capture and economic imbalance, particularly for those who rely on an open, innovative, and competitive internet.

At the Coalition for Creativity (C4C), our members span libraries, research institutions, digital rights organisations, technology businesses, and cultural heritage organisations. And we are alarmed by proposals that could subtly, yet significantly, reshape the rules of internet engagement.

When “Cooperation” Means Compulsion

One of the most concerning elements is the Commission’s suggestion to give new powers to national regulators and the Body of European Regulators for Electronic Communications (BEREC) to “facilitate cooperation” between network operators and content/application providers (CAPs). That may sound collaborative. In practice, it opens the door to mandated payments from CAPs to telecom providers, a model called “sending party pays”, and unsuccessfully trialled in, for example, South Korea.

Let’s call it what it is: a toll on digital expression.

There is no market failure here to solve. BEREC’s own findings repeatedly confirm that the IP interconnection market is competitive and stable. Prices are falling. Network efficiency is increasing. So why reengineer a system that functions?

The danger isn’t just financial. It’s structural. A new payment regime could disproportionately burden smaller players—research platforms, non-profit tools, cultural digitisation projects—who lack the resources to absorb unpredictable and potentially arbitrary costs. It would force many to divert resources away from innovation, access, and public service into opaque negotiations or legal defences.

From Enforcer to Arbitrator: The Risk of Regulatory Weaponisation

Granting regulators the mandate to broker financial flows between CAPs and network operators also introduces a more insidious risk: regulatory weaponisation.

We’ve already seen how dominant telecom operators can manipulate regulatory ambiguity to their advantage. Prolonged disputes, threats of complaints, or pressure through access negotiations become tools not of dialogue, but of dominance. The eleven-year legal battle between Init7 and Swisscom is a sobering illustration of how long and costly such “cooperation” can become when left unchecked.

Once financial mediation becomes part of a regulator’s role, it risks distorting its core mission—protecting end-users, ensuring market fairness, and safeguarding fundamental rights. Instead, it creates an uneven playing field where size and political access—not public interest—determine who pays, who wins, and who gets squeezed out.

Clarifying Net Neutrality, or Carving Loopholes?

The second DNA proposal promises to “clarify” Open Internet rules for “innovative services.” But clarity can sometimes be a pretext for reinterpretation. The vague invocation of “innovation” risks being exploited as a backdoor for paid prioritisation, “specialised services,” and network slicing models that would allow telecoms to tier the internet.

In such a system, services that can afford to pay for “fast lanes” will reach users quickly and reliably. Those that cannot—open-source educational tools, digitised cultural archives, niche research platforms—may be relegated to slower, less stable delivery. What we risk is a digital hierarchy, where value is dictated not by societal impact or user need, but by the capacity to pay for preferential treatment.

Net neutrality was designed to prevent this. It ensures that traffic flows equally, irrespective of origin. It protects the small, the emerging, the non-commercial. Diluting this principle under the guise of “interpretative guidance” would fundamentally alter the nature of European internet access.

The True Meaning of a Level Playing Field

A genuinely level digital playing field is one where access is open, terms are transparent, and opportunities are not auctioned off to the highest bidder. It is one where libraries, start-ups, research institutions, and cultural projects can reach users on the same terms as commercial giants.

What it is not: a market where incumbent telecoms dictate who pays, how much, and for what. It is not a network that throttles diversity in the name of “cooperation.” And it is certainly not a regulatory model where financial “facilitation” becomes the norm rather than the exception.

Our Ask to the Commission

C4C urges the Commission to:

1. Reject any mechanism that leads to network usage fees. The current IP interconnection market does not require this intervention.
2. Reinforce—not reinterpret—net neutrality. Paid prioritisation, however dressed up, must remain off the table.
3. Maintain regulatory independence. The role of NRAs and BEREC should be to enforce the law, not to arbitrate payments.
4. Base decisions on independent evidence. Not on narratives of scarcity pushed by a few powerful stakeholders.

A Pivotal Moment for Europe’s Digital Future

The DNA could set the tone for a new phase of European digital policy. It could reaffirm openness, diversity, and competition. Or it could signal a shift toward pay-to-access models that undermine the very fabric of our digital society.

Let’s not be lulled by comforting phrases or abstract frameworks. Let’s focus on outcomes—and on the users, creators, educators, and researchers who make Europe’s digital ecosystem one of the most vibrant in the world.

Because a “level playing field” that tilts against openness isn’t level at all.

For detailed insights, see our full submission.

Cross-posted from C4C’s LinkedIn page – see the original article

In late January, the Motion Picture Association (MPA) co-organised an exclusive screening in Brussels of Brady Corbet’s movie ‘The Brutalist’, which was awarded three Golden Globes.

Budapest, the Hungarian capital, was the prime filming location for this movie, which stars Adrien Brody as a Hungarian-born Jewish Holocaust survivor who emigrated to the US. Zoltán Tarr, a Hungarian Member of the European Parliament (MEP) who is the European People’s Party (EPP) coordinator for the EP’s Culture and Education (CULT) Committee, delivered the opening remarks at this Brussels screening.

The film’s editor, Dávid Jancsó, discusses AI’s role in the (post-)production process in an interview with RedShark Media, which explores the technical intricacies of making this movie. This article set off a snowball effect, creating controversy about the movie’s Oscar-worthiness for its use of AI, to the point that Corbet felt the need to issue a reaction to Deadline.

What’s the Fuss About?

A Ukrainian AI tool, Respeecher, was used to perfect the Hungarian pronunciations in some dialogues.

Jancsó, a native Hungarian speaker, considers it “one of the most difficult languages to learn to pronounce” and points out that the actors did “a fabulous job”. Then why the turn to AI? It was to “perfect it so that not even locals will spot any difference,” he explained.

“It is controversial in the industry to talk about AI, but it shouldn’t be,” Jancsó said, as he reminded, “There’s nothing in the film using AI that hasn’t been done before. It just makes the process a lot faster”. Deadline makes an interesting observation in this context: “We understand that the use of AI on The Brutalist didn’t eliminate jobs, rather created them.”

Garbage Out Due to Garbage In Is Likely to Affect EU Quality Content Most

Putting aside all the controversy around human creativity and the use of AI, this debate highlights broader policy implications. Using AI tools for applications in local languages requires that these models ingest quality training data in those languages to avoid a “garbage in, garbage out” scenario.

This raises an important question: What if quality Hungarian audio data had not been available for training AI tools like Respeecher?

If AI models lack diverse and representative linguistic data, the ability to refine pronunciation and enhance linguistic authenticity in films diminishes. This directly affects how local languages—especially those with smaller global reach—are represented in creative productions. The fact that The Brutalist benefited from AI-assisted pronunciation refinement suggests that Hungarian speech data was accessible for AI training. However, restrictive policies that limit the availability of local-language content for AI model training could jeopardize future applications of these technologies, to the detriment of the production of content in local languages.

This case exemplifies a broader reality: discussions about AI should not just be about the potential threat to the creative sector—we need to acknowledge AI is also a powerful enabler of authenticity, efficiency, and new creative possibilities. The challenge ahead for policymakers is not to block AI’s access to cultural and linguistic diversity but to ensure that AI systems are built in a way that respects creators while enabling innovation.

If we strip AI of the ability to learn from local content, the first victim may well be the very authenticity and cultural richness the EU seeks to preserve.

Cross-posted from C4C’s LinkedIn page – see the original article

Caroline De Cock, Coordinator of the Coalition for Creativity (C4C), recently had the opportunity to speak at the EU general-purpose AI Code of Practice plenary meeting of Working Group 1 – Copyright. She addressed key concerns about the intersection of copyright and artificial intelligence (AI). Her intervention highlighted the need for a balanced approach that fosters innovation, respects existing laws, and avoids overreach. Below is a summary of the key points she shared.

Copyright and the Code of Practice: Drawing the Line

At its core, the Code of Practice must respect existing boundaries—it should not expand copyright law or impose obligations beyond those established by the AI Act. Any uses that do not infringe copyright must remain outside its scope. Equally important is preserving the critical role of copyright exceptions and limitations, which are essential for research, education, and creativity.

Transparency and Due Diligence: Complex Ideals in an Opaque Environment

While transparency is vital, it must also be practical. Copyright law is notoriously complex, with low originality thresholds, fragmented territorial rules, and no universal registration system. Metadata on rights ownership is often incomplete or missing. In such an environment, requiring granular transparency or extensive due diligence is simply unworkable. The Code of Practice must recognise these limitations to ensure that compliance is achievable.

Opt-Outs: Respect Without Overreach

Opt-out mechanisms, such as the Article 4 reservation of rights in the 2019 Copyright Directive, play a role in AI’s interaction with copyright. However, their current implementation raises unresolved questions:

• How should opt-outs be expressed?
• How can their legitimacy be verified?

These challenges remain a work in progress and should not be preempted by the Code of Practice. Additionally, an opt-out represents a reservation of rights, not an outright prohibition. For instance, lawful uses permitted by copyright exceptions and limitations should remain valid even when an opt-out is in place.

Broadening the Focus: Beyond Entertainment

Generative AI offers transformative potential across sectors, yet discussions around General-Purpose AI (GPAI) often focus narrowly on the entertainment industry. This risks neglecting other critical domains, such as research and education, which are pivotal to societal progress. Unfortunately, voices from the research community remain underrepresented in the development of this Code. To achieve fairness and inclusivity, we must ensure that their needs, particularly regarding access to knowledge, are not overshadowed.

Learning from Article 17: The Danger of Output Filters

A major concern is the potential for Measure 2.9 of the Code to introduce “output filters”—tools that could block AI-generated content based on copyright concerns. This is an alarming step beyond the upload filters debated under Article 17 of the 2019 Copyright Directive. Unlike upload filters, which apply to public contexts, output filters would target private, exploratory uses of AI, effectively introducing pre-emptive censorship of thought and experimentation. This poses a serious threat to freedom of thought and expression in the EU and must be avoided at all costs.

A more detailed overview of our concerns on this point can be found in this Tech Policy Press article.

Conclusion: Striking the Right Balance

As we navigate the rapidly evolving relationship between copyright and AI, it is imperative to create a framework that is balanced, practical, and inclusive. This means respecting the boundaries of existing laws, ensuring feasible transparency, and broadening the discussion beyond entertainment to include sectors like research and education.

Let’s continue this critical conversation to shape an AI framework that serves everyone.

Get the TL;DR version 👆

Cross-posted from C4C’s LinkedIn page – see the original article

The Coalition for Creativity (C4C) recently submitted its comments to the Irish Department of Enterprise, Trade and Employment on the evolving relationship between generative AI, copyright, and related rights. As a diverse coalition representing libraries, research institutions, digital rights advocates, and technology businesses, our mission is to ensure that copyright frameworks foster innovation, access, and creativity while supporting creators.

Here are some key insights from our submission:

1. Balancing Copyright and Innovation: While the EU legal framework, especially the exceptions for text and data mining (TDM), has provided much-needed legal clarity, there’s still a critical need for balance. Restricting AI developers to public domain datasets could lead to biased outputs and reduced innovation. We advocate for copyright policies that encourage responsible AI development by ensuring access to diverse datasets while allowing creators to safeguard their rights.
2. Practical and Granular Tools for Opt-Out Mechanisms: Current opt-out tools are a positive start but lack the granularity needed for today’s complex AI environment. We recommend introducing more refined mechanisms that differentiate between full opt-outs (e.g., for TDM) and opt-outs specific to generative AI training. Importantly, any new policies must avoid retroactively applying opt-out decisions to models that have already been trained.
3. Protecting Quality and Cultural Diversity: A potential “data winter” is already on the horizon if access to diverse datasets is unduly restricted. This not only risks stifling innovation but also threatens the cultural richness that AI tools can reflect. A wide variety of datasets is essential to ensure fairness and avoid biases in AI-generated content, which is why we stress the importance of policies that support both creativity and diversity.
4. Copyright and AI-Generated Works: The question of whether AI-generated works should qualify for copyright protection is a complex one. C4C’s position is that any copyright protection should depend on the level of human creative intervention. Extending copyright to fully AI-generated works runs counter to the core principle of copyright, which is to incentivize human creativity.
5. Opposing New Sui Generis Rights: We strongly oppose the introduction of additional sui generis rights for AI-generated content. The experience with the EU Database Directive has shown that such rights often create unnecessary legal complexities and hinder rather than help innovation. A balanced framework, not additional layers of proprietary rights, is the way forward.
6. Ensuring Fair Access to Data: Lastly, it’s crucial that smaller AI developers are not locked out of access to high-quality data. Disproportionate restrictions or overly burdensome licensing regimes would tilt the playing field in favor of larger developers and could lead to a monopolization of AI development. Policies should ensure that diverse datasets are accessible, particularly for smaller entities that drive much of the innovation in AI.

Conclusion

As AI continues to reshape industries and society, the copyright frameworks governing its use must remain flexible and forward-thinking. C4C is committed to advocating for policies that protect creators while also ensuring that AI can thrive as a tool for creativity, innovation, and public benefit.

For detailed insights, see our full submission.

Cross-posted from C4C’s LinkedIn page – see the original article

The Coalition for Creativity (C4C) submitted its response to the European Commission’s White Paper on “How to Master Europe’s Digital Infrastructure Needs.” Our feedback focuses on several critical areas where we believe the Commission’s proposals require reconsideration. You can contribute too, until 30 June!

Here’s a summary of our key points:

1. Clarifying convergence vs. coexistence: The White Paper’s push for regulatory convergence of cloud and telecommunications services is based on a misconception. These sectors operate under distinct market conditions and necessitate separate regulatory frameworks. A forced convergence could stifle innovation and disrupt market dynamics.
2. Real level playing field: Creating a level playing field requires recognising the fundamental differences between telecoms and cloud providers. The hypothetical convergence suggested by the White Paper doesn’t align with market realities. Regulation should respond to existing market dynamics, not theoretical scenarios.
3. Regulatory balance across the Internet stack: The narrative that cloud providers escape stringent regulations is misleading. Cloud services are governed by robust regulatory frameworks covering consumer protection, security, and interoperability. Extending telecom-specific regulations to these sectors could hamper competition and innovation, and simply does not make sense.
4. Questionable indirect contribution models: Proposing indirect contributions through a universal service obligation after the strong/forceful pushback against network fees risks introducing inefficiencies and regulatory burdens. This approach could distort market incentives, increase costs for all market players and consumers, and deter timely investments in network infrastructure.
5. Unnecessary dispute resolution mechanisms: The White Paper’s proposal for new dispute resolution mechanisms lacks justification, as BEREC has repeatedly noted the absence of any substantial legal or economic evidence to support regulatory intervention in the peering market. Introducing unnecessary regulatory layers could create more problems than it solves.

Our Recommendations:

• Focus on market-driven integration: The Commission should prioritise creating a regulatory environment that respects the distinct needs of each sector, fostering integration driven by market forces.
• Maintain pro-competitive frameworks: Stick to the original goals of the telecoms regulatory framework, reducing ex ante regulation where competition laws suffice.

Our Words of Caution:

• Avoid industrial policy interventions: Legislative approaches should be evidence-based, and avoid disrupting well-functioning markets such as the Internet ecosystem.
• Reject network fees: The concept of network fees has been widely rejected and should not be pursued further, as it only benefits a select few at the whole ecosystem’s expense.

Conclusion

In conclusion, C4C believes that the European Commission’s White Paper predominantly reflects aspirational goals rather than the current market realities. It presupposes a convergence between computing and connectivity sectors that does not yet exist, and may not align with customer or developer preferences. This approach risks altering the Internet’s inherent structure through regulatory means without genuine market demand.

For detailed insights, see our full submission.

Cross-posted from C4C’s LinkedIn page – see the original article

Note: Trying to turn a cat around by its tail is a typical Polish expression (Odwracać kota ogonem) that refers to the gentle art of presenting facts in a false or distorted light.

On 15 February, the Polish government published its DCDSM transposition proposal (draft act – explanatory memorandum). The text is open to comments for 14 days after its publication date.

C4C is concerned about the non-compliance of the Polish implementation approach regarding:

1. Text and Data Mining (TDM) and generative AI; and,
2. the use of works for educational and scientific purposes.

In this context, we want to share our draft comments below early on to help raise broader awareness about these two worrying elements.

1) TDM and generative AI

The Polish implementation approach considers that the training of generative AI does not fall within the scope of the DCDSM TDM provisions (Articles 3 and 4). It introduces a carveout to exclude the application of Articles 3 and 4 to generative AI models. To justify this, it claims in its explanatory memorandum that the effects of generative AI could not have been foreseen at the time of the Directive’s adoption in 2019.

C4C believes that excluding from the scope of the TDM provision the reproductions made in the context of training generative AI models would clearly result in a non-compliant implementation.

Defined in Article 2(2) as “any automated analytical technique aimed at analysing text and data in digital form in order to generate information which includes, but is not limited to, patterns, trends and correlations”, the term must be considered as an autonomous concept of EU law that cannot be modified by Member States.

The argument raised in the explanatory memorandum that “it seems fair to assume that this type of permitted use was not conceived for artificial intelligence,” is incorrect, as evidenced by European Commissioner Breton’s acknowledgement, in response to a Parliamentary question, that the DCDSM “introduces exceptions covering text and data mining (TDM) that are relevant in the AI context” and that “these exceptions provide balance between the protection of rightholders including artists and the facilitation of TDM, including by AI developers.”

Moreover, TDM is about mining, rather than the outcomes of the process, and Artificial Intelligence was already a reality in the fringes of the DCDSM discussions, albeit mostly at the business-to-business level.

Paragraph (13a) of the Council’s agreed negotiating mandate of 25 May 2018 (9134/18) stresses that “text and data mining techniques are widely used both by private and public entities to analyse large amounts of data in different areas of life and for various purposes, including for government services, complex business decisions and the development of new applications or technologies”, leading it at that stage of the negotiations to ask for the introduction of an optional exception (Article 3a) for enabling public and private entities to use mining techniques to access data which are lawfully accessible (for instance when they are freely available to the public online)”.

Finally, the recently adopted AI Act explicitly confirms the interplay with the DCDSM, and more specifically with its TDM provisions.

C4C moreover considers the arguments to be additionally flawed at various levels:

1. Technology-neutral framework of the Directive: The DCDSM is designed with a technology-neutral framework to ensure its applicability over a wide range of digital activities and technologies, including those not explicitly foreseen at the time of its drafting. This approach reflects an understanding that the digital environment is rapidly evolving, and laws must be adaptable to future developments.
2. Principle of dynamic interpretation in line with evolving technologies: The Directive’s interpretation should evolve alongside technological advancements to fulfil its overarching goals, which include protecting copyright holders in the digital age and promoting a balanced and fair digital market. The dynamic interpretation of legal texts allows for the inclusion of technologies such as generative AI, which, although not explicitly mentioned, fall within the scope of activities impacting copyright and the sharing of digital content.
3. Purpose and objectives of the Directive: The primary objectives of the Directive include harmonising and modernising the rules applying to copyright and fostering a healthy digital environment for innovation and creativity. Generative AI falls within the ambit of the Directive, aligning with its purpose of adapting copyright rules to the digital age.
4. Legislative intent and evolving digital landscape: Lastly, considering the legislative intent behind the Directive to enable a digital single market and the evolving nature of the digital landscape, it is reasonable to assert that the Directive’s provisions apply to generative AI. The law must be responsive to changes in how content is created, distributed, and consumed, to ensure that copyright protections remain effective and relevant.
5. Interpretative flexibility of EU law: EU law is subject to interpretation by the Court of Justice of the European Union (CJEU), which has a history of interpreting directives in a manner that takes into account the evolving digital landscape and the objectives of the EU legislation. This means that even if a specific use case was not explicitly considered during the legislative process, it does not preclude its inclusion under the scope of the directive if such inclusion serves the directive’s objectives.

2) The use of works for educational and scientific purposes

The transposition proposal introduces a new restriction on using works for educational and scientific purposes, limiting it to 20% of the volume, which is also not in line with the exception as set out in the DCDSM and hence constitutes an infringement.

Featured image created with DALL·E