EUVIMEDEUVIMED entdecken
Crossref

Copia negata: training datasets as essential facilities

Sudhanwa Sandeep Joshi

Competition Law Journal · 2026-11 · 25 (2) · 117-124

DOI: 10.4337/clj.2026.02.04 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Competition Law Journal, 2026-11. Verfasst von Sudhanwa Sandeep Joshi. Die Arbeit nennt 71 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Where the provenance of a training corpus determines the capability of the resulting model, an undertaking controlling a historically accumulated and non-replicable dataset holds an input its downstream rivals cannot obtain elsewhere. An increasing number of such undertakings have reserved their corpora through various means while entering downstream artificial intelligence markets themselves. This article considers whether the essential facilities doctrine under Article 102 TFEU reaches refusals to license training data and argues that the question resolves along two routes. Where a corpus is withheld from all comers, the classical Bronner conditions apply, and indispensability is established through historical non-replicability. Where a corpus is licensed generally but withheld for artificial intelligence training, Android Auto displaces the requirement for indispensability in favour of an effective-competition standard. The market definition difficulty that has obstructed data-as-facility claims dissolves once a training corpus is understood as a composite input whose layers answer to different definitions. However, this possibility tests the previously defined contours of competition law vis-à-vis intellectual property rights, as what a model reproduces from its training data remains unresolved: the Court of Justice is seized of one half of the problem in Like Company v. Google , and, of the other half, no court is seized at all. Compelling access thus risks dissolving the incentive to assemble corpora in a manner which no prior essential facilities claim has had to confront, and the article seeks to ensure that this dilemma is considered by regulators and courts when such litigation arises.

Zitieren

APA

Joshi, S. S. (2026). Copia negata: training datasets as essential facilities. Competition Law Journal, 25(2), 117-124. https://doi.org/10.4337/clj.2026.02.04

Vancouver

Joshi SS. Copia negata: training datasets as essential facilities. Competition Law Journal. 2026;25(2):117-124. doi:10.4337/clj.2026.02.04

BibTeX
@article{Joshi2026,
  author = {Joshi, Sudhanwa Sandeep},
  title = {Copia negata: training datasets as essential facilities},
  journal = {Competition Law Journal},
  year = {2026},
  volume = {25},
  number = {2},
  pages = {117--124},
  doi = {10.4337/clj.2026.02.04}
}
Kostenlos starten →