Copyright and Training Generative Models: Content Transparency and Independent Auditing

Copyright and Training Generative Models: Content Transparency and Independent Auditing

Published
November 19, 2026
Author
Khrabrykh S. A.
Original language
English

The report examines the legal and evidentiary chain involved in using protected content to train generative models. Its starting point is that the legality of training is not determined by a single “artificial intelligence” permission: it depends on rights in works and databases, lawful access, the applicability of text and data mining (TDM) exceptions, respect for rights reservations, data protection and the model’s behavior after release.

The analysis covers EU law as of 23 October 2026: Articles 3 and 4 of the DSM Directive, the French transposition in Article L122-5-3 of the CPI, the obligations of GPAI model providers under Article 53 of the AI Act, and the voluntary GPAI Code of Practice. The comparative part considers fair use and the Bartz and Kadrey cases in the US, the outcome of the UK consultation of March 2026, Singapore’s rules on computational data analysis, and a first-instance case in Munich concerning song lyrics. Separate sections address opt-out mechanisms, chain of title, licensing, memorization and digital evidence.

The central conclusion is that neither a public URL, nor a statement about the use of “public sources”, nor an Article 53 summary confirms the lawfulness of each work included. The report distinguishes binding rules, official guidance, judicial assessments in specific cases and research proposals, and warns against presenting first-instance rulings or settlements as universal rules.

For Observatoire ARGA, the report proposes an independent verification model with five levels: a legal map of rights and exceptions; a provenance register for source objects and derivative files; tests of how machine-readable restrictions are handled; tests of the risk of reproduction in model outputs; and public reporting that states the limits of uncertainty. It also outlines possible ARGA positions for the European Commission’s Technology and Copyright consultation, from a minimum set of provenance fields to low-cost dispute procedures for small rightsholders, and includes 64 appendices with forms, protocols and hypothetical training cases.

The authors stress that the audit methodology is a proposal, not a legally established certification. The report does not establish infringement by any specific company, is not a conformity assessment of their models and does not in itself constitute a consultation submission; the appendices are not the result of an external audit actually carried out.

Full text of the document
ARGA Observatory

Citation Rules

  1. Mandatory source attributionWhen using ARGA Observatory materials, the full name must be cited.
  2. Date and version indicationFor analytical reports, the year of publication is mandatory.
  3. Link to the originalElectronic materials must be accompanied by an active link to the official website.
  4. Context preservationCitations must not be shortened or altered in a way that distorts the original meaning.
  5. Note on adaptationIf the text is abridged or translated, state: "adapted from ARGA Observatory report".
  6. No commercial use without written permission from the organization
  7. Data accuracy preservationCharts and tables must be reproduced without changes.
Confidential enquiry

Ask the ARGA experts

Four short steps. We reply within one business day.

Step 1 of 4
What situation are you facing?
What stage is the case at?
Describe the situation
You can skip this field.
How should we reach you?
Your enquiry goes straight to ARGA experts. We do not publish it or pass it to third parties.
Official registration
ARGA in international registries
UNGM ID
1232417
ARGA is registered in the United Nations procurement system.
View registration
UN Global Compact
213173
Participant of the UN Global Compact.
ADB GMS Number
049963
Asian Development Bank — registered participant.
Scroll to Top