Art Newspaper argues unchanged AI output negates infringement

The Art Newspaper published an analysis on September 4, 2026, arguing that AI-generated images may not constitute copyright infringement if removing a specific copyrighted work from training data produces no change in the output. The publication posits that this functional test could serve as a valid legal defense by demonstrating the absence of causal reliance on the plaintiff's work. This interpretation suggests courts might evaluate infringement based on generative results rather than mere presence in training datasets. Legal experts note this argument challenges prevailing theories that focus exclusively on unauthorized ingestion during model development. The analysis cites ongoing litigation where defendants seek to distinguish between data exposure and expressive copying. If adopted judicially, this standard would require plaintiffs to prove specific causal links between their works and generated outputs. Critics warn this approach ignores the collective value of aggregated training data even when individual contributions appear statistically negligible.