Can GPAI providers use any publicly available content for training?

Not unconditionally. Under Article 4 of Directive 2019/790, rights holders may opt out of text and data mining by expressing reservations in a machine-readable format. GPAI providers must respect these opt-outs as required by both the Copyright Directive and Article 53(1)(c) of the AI Act.

What format must the training data summary take?

The AI Office is developing a template for the training data summary required under Article 53(1)(c). The summary must be sufficiently detailed for rights holders to understand whether their works may have been included in training data, without requiring disclosure of every individual work.

Does the copyright obligation apply to models trained before the AI Act?

GPAI models placed on the EU market after the compliance deadline must meet all Article 53 obligations, including copyright policy requirements, regardless of when training occurred. Providers of existing models should audit their training data practices and establish compliant policies.

How should providers handle robots.txt for AI training?

Providers should implement crawling infrastructure that checks robots.txt directives related to AI training and TDM before processing any content. Regular re-scanning of previously crawled sources is recommended to capture updated opt-out signals.

Can a provider face both copyright and AI Act penalties for the same violation?

Yes. Copyright infringement is enforced through national courts, while AI Act violations are enforced by the AI Office. A provider failing to respect copyright opt-outs could face both copyright damages and administrative fines of up to EUR 15 million or 3% of annual worldwide turnover under the AI Act.

Quick answer

GPAI model providers must comply with EU copyright law, including the text and data mining provisions of Directive 2019/790. Under Article 53(1)(c) of the EU AI Act, providers must put in place a policy to respect opt-out reservations expressed by rights holders under Article 4 of the Copyright Directive, and must provide a sufficiently detailed summary of training data content.

Updated June 2026 · MmowW AI Compliance

GPAI Copyright Compliance Under the EU AI Act (2026) | MmowW

The Intersection of AI and Copyright Law in the EU

The EU AI Act creates a direct link between GPAI model provision and copyright compliance. Article 53(1)(c) requires providers of general-purpose AI models to put in place a policy to comply with Union copyright law, in particular to identify and comply with reservations of rights expressed by rights holders pursuant to Article 4(3) of Directive 2019/790. This obligation applies to all GPAI model providers making models available on the EU market.

This requirement reflects the EU's recognition that GPAI model training typically involves processing large volumes of copyrighted material, and that rights holders must retain meaningful control over how their works are used in this context.

EU Copyright Directive: Text and Data Mining Framework

Article 3: Research Exception

Article 3 of Directive 2019/790 provides a mandatory exception for text and data mining (TDM) carried out by research organisations and cultural heritage institutions for scientific research purposes. This exception cannot be overridden by contract and applies regardless of whether rights holders have expressed any reservation. However, this exception is narrowly scoped and does not apply to commercial GPAI model training.

Article 4: General TDM Exception and Opt-Out

Article 4 of Directive 2019/790 provides a broader TDM exception that allows any party, including commercial GPAI providers, to mine lawfully accessible works. However, this exception is subject to a critical condition: rights holders may reserve their rights by opting out of TDM. For online content, this reservation must be expressed in a machine-readable format, such as through robots.txt directives or metadata tags.

When a rights holder has expressed such a reservation, GPAI providers must respect it. Article 53(1)(c) of the AI Act transforms this copyright obligation into an AI regulatory requirement, meaning that failure to respect opt-outs is not only a copyright infringement but also an AI Act compliance violation.

Practical Compliance Requirements for GPAI Providers

Establishing a Copyright Policy

GPAI providers must establish and maintain a documented copyright compliance policy. This policy should address how the organisation identifies copyrighted content in training data, how it detects and respects opt-out reservations, and what processes are in place to handle rights holder inquiries and complaints.

The policy should be made available to the AI Office upon request and should be referenced in the provider's technical documentation under Annex XI.

Detecting and Respecting Opt-Outs

Providers must implement technical measures to detect opt-out reservations expressed by rights holders. For web-crawled data, this typically involves respecting robots.txt directives that specifically address AI training or TDM activities. Providers should also monitor metadata standards and industry protocols for expressing TDM reservations.

Key implementation steps include:

Crawling infrastructure that checks robots.txt and relevant metadata before processing content
Regular re-scanning of previously crawled sources for updated opt-out signals
Maintaining records of opt-out detections and actions taken
Establishing a contact point for rights holders to communicate reservations directly

Training Data Transparency

Article 53(1)(c) also requires providers to make publicly available a sufficiently detailed summary of the content used for training the GPAI model. This summary must be drawn up according to a template provided by the AI Office and should enable rights holders to understand whether their works may have been used in training.

The summary does not require listing every individual work used, but must provide meaningful information about the types of content, sources, and volumes of data processed during training.

Relationship Between Copyright and AI Act Enforcement

Copyright compliance under the AI Act is enforced through two parallel tracks. Copyright infringement claims remain within the jurisdiction of national courts and copyright enforcement bodies under existing frameworks. However, failure to comply with the AI Act's copyright policy requirements falls under the jurisdiction of the AI Office.

This means that a GPAI provider could face both copyright infringement proceedings and AI Act administrative penalties for the same underlying conduct. Administrative fines under the AI Act for non-compliance with Article 53 obligations can reach EUR 15 million or 3% of annual worldwide turnover.

Emerging Challenges and Open Questions

Several aspects of GPAI copyright compliance remain subject to ongoing interpretation. The definition of machine-readable opt-out remains an evolving standard, with different platforms and rights holders adopting varying technical approaches. The scope of the training data summary requirement is being refined through the GPAI Code of Practice process.

Additionally, questions about retroactive application to models trained before the AI Act's entry into force, and about the treatment of data obtained through third-party datasets, continue to be discussed by regulators and stakeholders.

Preparing for Compliance

Organisations should begin documenting their copyright compliance practices now, well ahead of the August 2026 deadline. This includes auditing current training data pipelines for copyright compliance, implementing opt-out detection mechanisms, and preparing the required training data summary.

Start your AI compliance journey to track your copyright compliance readiness alongside other GPAI obligations.

Check your AI compliance readiness — free.

Take the Readiness Check 3 minutes · 10 questions · no signup required

This article is for informational purposes only and does not constitute legal advice. Regulatory requirements change frequently — verify current rules with official sources. Built by Sawai Gyoseishoshi Office, Hiroshima, Japan.