When the Opt-Out Fails: Unfair Competition Law and the Control of AI Training Inputs in the EU

The text-and-data-mining opt-out in Art. 4(3) of the Digital Single Market Directive is widely treated as the cornerstone of rightholder control over the use of protected works in artificial intelligence (AI) training. This article argues that the opt-out is structurally unable to deliver that control, because its effectiveness depends on a precondition that modern training pipelines routinely defeat, namely the traceability of a work’s provenance. Once provenance is diluted or erased, a formally valid reservation of rights becomes practically unenforceable. The transparency obligations introduced by Art. 53(1)(c) and (d) of the Artificial Intelligence Act (AI Act) sharpen the duty but do not close the gap, since the training-data summary is designed to be comprehensive rather than granular. Building on this analysis, the article contends that unfair competition law offers a control mechanism that copyright alone can no longer guarantee. While existing scholarship has applied unfair competition reasoning to AI output, this article turns it to the input side, reframing opaque data sourcing and the circumvention of machine-readable reservations as unfair commercial conduct measured against the transparency benchmark set by the AI Act itself. Anchored primarily in EU law, with Spanish unfair competition law as a national reference, it presents unfair competition not as a substitute for copyright but as a complementary bridge restoring meaningful control over AI training inputs.