Targeted initiative for a better copyright environment for European creativity and innovation

AI Chamber submitted its position in the European Commission’s consultation on the “Targeted initiative for a better copyright environment for European creativity and innovation”. In the position, we highlight the need to protect innovation, legal certainty and proportionate rules on AI, TDM, licensing and the protection of creators.

1. Introduction

The AI Chamber welcomes the opportunity to provide feedback on the Commission’s combined Call for Evidence concerning the review of Directive (EU) 2019/790 on copyright and related rights in the Digital Single Market and the possible targeted legislative initiative for a better copyright environment for European creativity and innovation.

The AI Chamber is a business association supporting the responsible development, deployment and adoption of artificial intelligence in Central and Eastern Europe. Our mission is to connect innovators, policymakers, researchers and business leaders, and to ensure that the voice of small and medium-sized enterprises from the CEE region is heard in EU and national policy processes. We primarily represent SMEs that build, use or implement AI technologies in Central and Eastern Europe, including start-ups, research-oriented businesses, applied AI companies, consultancies and technology providers.

This perspective is essential. The copyright framework for AI training will not affect all market actors equally. Large global technology companies may be able to absorb complex licensing negotiations, extensive documentation duties and litigation risk. European SMEs, start-ups, open-source developers and research-driven scale-ups often cannot. For them, legal uncertainty or disproportionate transaction costs may prevent market entry entirely.

AI Chamber therefore supports a balanced review of the CDSM Directive, but strongly cautions against any reform that would narrow the existing text and data mining framework or create de facto mandatory licensing for AI training.

2. Executive summary

AI Chamber CEE submits that the Commission should:

1. preserve, clarify and strengthen Articles 3 and 4 of Directive (EU) 2019/790, rather than narrow them;
2. avoid mandatory licensing, ex ante authorisation or remuneration schemes for AI training based on lawfully accessible content, especially where such schemes would impose disproportionate burdens on SMEs;
3. reject duplicative copyright-specific transparency duties that overlap with the AI Act, in particular Article 53 obligations for general-purpose AI models;
4. ensure maximum harmonisation of the TDM and scientific research exceptions to prevent Member State fragmentation and gold-plating;
5. apply an SME and innovation impact test before proposing any new copyright obligation affecting AI development;
6. improve the research and secondary publication framework, including by enabling researchers to commercialise publicly funded research outputs without legal uncertainty;
7. address performer impersonation, deepfakes and personality-right concerns through targeted, proportionate rules, not through an overbroad restriction of access to data for AI training.

3. The consultation must not treat AI primarily as a copyright enforcement problem

AI Chamber recognises that generative AI raises legitimate questions for rightholders, performers and creative sectors. We do not dispute the importance of fair copyright markets, effective enforcement against infringement, or appropriate remedies against unlawful impersonation and piracy.

However, the starting point of EU policy must be broader. AI is a general-purpose technology. It is relevant not only to cultural and creative industries, but also to healthcare, energy, climate technologies, cybersecurity, logistics, public administration, industrial optimisation, accessibility tools, language technologies and scientific discovery. Many AI systems depend on access to large-scale data but do not compete with creative works, do not substitute protected expression, and have no meaningful connection to the exploitation of cultural content.

A copyright reform designed around the most contested examples of generative AI would therefore risk capturing many socially beneficial and economically critical AI uses. This would be especially damaging for SMEs, which often develop narrow, applied, domain-specific AI systems rather than large consumer-facing generative models.

The Commission should therefore distinguish clearly between:

• unlawful reproduction or communication of protected expression;
• lawful text and data mining of lawfully accessible material;
• downstream infringing outputs;
• personality-right or consumer-protection issues such as deceptive impersonation;
• online piracy of live or time-sensitive content;
• contractual disputes about remuneration in specific creative sectors.

These issues should not be collapsed into a single licensing-focused response.

4. Articles 3 and 4 CDSM are the legal foundation for European AI development

Text and data mining is not a marginal activity. It is a core technical process for modern AI development, data analytics, scientific research and machine learning. Articles 3 and 4 of Directive (EU) 2019/790 provide the legal basis that enables EU actors to conduct TDM under defined conditions.

For SMEs, this framework is indispensable. It allows innovators to build and test models without negotiating licences for every item of lawfully accessible material used in computational analysis. The current framework also reflects a reasonable balance: Article 3 protects scientific research, while Article 4 establishes a broader TDM exception subject to rights reservations.

AI Chamber submits that the Commission should not narrow Articles 3 and 4, carve out AI training from their scope, or introduce a special authorisation layer for machine learning. Such a move would significantly increase legal risk and transaction costs for European developers and would likely favour the largest market incumbents.

Instead, the Commission should clarify that reproductions and extractions necessary for TDM and machine learning on lawfully accessible material fall within Articles 3 and 4, subject to the conditions already provided by the Directive, including valid rights reservations under Article 4(3).

5. Mandatory licensing would disproportionately harm SMEs

A shift from the current TDM framework towards mandatory licensing or quasi-mandatory remuneration would create substantial economic and legal barriers.

For SMEs, the main problem is not only the licence fee. It is the full transaction cost: identifying rightholders, verifying title, negotiating terms, documenting provenance, monitoring opt-outs, assessing national divergences, handling collective management complexity, and defending against litigation. These costs arise before a product is commercialised and before the company has generated revenue.

This is a structural disadvantage for European AI start-ups. Large foreign technology incumbents can internalise legal teams and compliance systems. SMEs cannot. A licensing-first model would therefore not necessarily increase fairness; it may entrench market concentration.

The Commission should avoid reforms that make lawful AI development dependent on access to complex licensing infrastructures controlled by established intermediaries. Any new measure must be tested against the practical ability of a small European AI company to comply without needing a multinational-level legal department.

6. The EU must avoid double regulation

The AI Act already establishes copyright-related obligations for providers of general-purpose AI models. In particular, Article 53 requires providers to put in place a policy to comply with Union copyright law, including respect for rights reservations under Article 4(3) CDSM, and to make publicly available a sufficiently detailed summary of content used for training.

This framework is still being implemented. It should be assessed empirically before additional sector-specific transparency or audit duties are introduced through copyright legislation.

AI Chamber therefore recommends that the Commission reject duplicative obligations such as:

• separate copyright training-data audits;
• model-by-model licensing disclosure requirements beyond the AI Act;
• ex ante approval mechanisms for datasets;
• obligations to disclose trade secrets or commercially sensitive dataset strategies;
• retroactive duties to remove, retrain or “machine unlearn” material where technically infeasible or economically disproportionate.

New copyright rules should be fully interoperable with the AI Act and should not create conflicting or cumulative compliance layers.

7. Legal certainty should be improved, not reduced

The current consultation identifies several real problems, including licensing challenges, piracy, remuneration and cross-border fragmentation. However, legal uncertainty should not be solved by imposing broader rights-control mechanisms before technical feasibility and economic impact are established.

For AI developers, uncertainty itself is a barrier to investment. If companies face open-ended liability, unclear opt-out standards, unpredictable remuneration claims or future retroactive obligations, capital and talent will move to jurisdictions with clearer rules.

AI Chamber therefore recommends the introduction of a clear EU-level legal certainty package:

1. a harmonised rule that TDM for machine learning on lawfully accessible material does not require additional authorisation, unless a valid rights reservation under Article 4(3) applies;
2. a clear, machine-readable and standardised opt-out mechanism;
3. protection against contractual override of the scientific research exception;
4. no retroactive liability for good-faith reliance on existing Articles 3 and 4;
5. proportionate remedies that do not require technically impossible or economically destructive retraining;
6. clear safe harbours for SMEs using documented compliance procedures.

8. Scientific research and secondary publication rights should be strengthened

AI Chamber supports the Commission’s attention to the fragmented implementation of the optional scientific research exception under Directive 2001/29/EC and the obstacles faced by researchers in sharing and reusing publicly funded research.

This issue is directly relevant to European AI competitiveness. Research results, datasets, articles and experimental outputs are essential inputs for AI development. Fragmented national exceptions create uncertainty for cross-border research collaboration and for research-driven commercialisation.

The EU should therefore:

• make relevant research exceptions mandatory and fully harmonised;
• ensure that researchers may engage in TDM and computational analysis without contractual override;
• create or strengthen secondary publication rights for publicly funded research;
• allow research organisations and researchers to commercialise research outputs without losing the benefit of lawful research-related data uses;
• support open science, open data and open-source AI ecosystems.

This approach would align copyright policy with the EU’s broader objective of technological sovereignty.

9. Performer impersonation should be addressed through targeted rules

AI Chamber recognises that AI-generated imitation of performers’ voices, likenesses or personal characteristics can raise serious concerns. However, these concerns are not always copyright issues. They may involve personality rights, consumer protection, unfair commercial practices, contract law, data protection or sector-specific performer protections.

The Commission should therefore avoid using performer impersonation as a justification for a broad restriction on AI training. Instead, it should consider targeted measures focused on deceptive commercial impersonation, unauthorised endorsement, fraud, reputational harm and unfair substitution in specific markets.

Such measures should be proportionate, technologically neutral and compatible with lawful uses such as parody, research, accessibility, security testing and non-deceptive creative tools.

10. Online piracy and RAAP-related remuneration issues should remain separate

AI Chamber does not oppose stronger, proportionate enforcement against online piracy of live events and time-sensitive content. However, anti-piracy measures should not become a vehicle for restricting lawful access to data or imposing general monitoring obligations on AI developers.

Similarly, the issue of single equitable remuneration for third-country phonograms following the CJEU judgment in RAAP is important, but legally and economically distinct from AI training. It should be addressed on its own terms and should not be merged with the TDM debate.

11. Maximum harmonisation and no gold-plating

Divergence between Member States is one of the greatest risks for SMEs. A start-up operating across the internal market cannot realistically manage twenty-seven different interpretations of TDM, opt-outs, research exceptions, remuneration claims and enforcement standards.

The CDSM review should therefore move towards maximum harmonisation in areas affecting AI development, especially:

• Article 3 TDM for scientific research;
• Article 4 TDM for general purposes;
• opt-out format and legal effect;
• contractual override;
• research and secondary publication rights;
• compliance expectations for SMEs.

Member States should not be allowed to introduce national carve-outs that narrow the practical availability of TDM exceptions or create additional licensing duties for AI training.

12. Recommendations

AI Chamber CEE recommends that the Commission include the following parameters in any forthcoming initiative:

First, preserve and strengthen Articles 3 and 4 of Directive (EU) 2019/790. AI training and machine learning should not be excluded from the scope of TDM.

Second, clarify that TDM on lawfully accessible material does not require additional authorisation, subject to valid and standardised rights reservations under Article 4(3).

Third, reject mandatory licensing or remuneration schemes for AI training that would impose disproportionate costs on SMEs.

Fourth, avoid double regulation with the AI Act. Copyright-specific transparency duties should not duplicate Article 53 AI Act obligations.

Fifth, establish a harmonised, machine-readable opt-out standard. Ambiguous website terms, non-standard notices or after-the-fact claims should not create legal uncertainty for developers.

Sixth, introduce SME-sensitive compliance standards, including simplified procedures, model clauses, safe harbours and proportionality safeguards.

Seventh, make research and secondary publication exceptions stronger, mandatory and cross-border.

Eighth, address performer impersonation and piracy through narrow, targeted instruments rather than broad restrictions on lawful data access.

Ninth, prohibit national gold-plating that fragments the internal market.

Tenth, conduct a dedicated innovation and SME impact assessment before proposing any new obligation affecting AI training.