The Stanford Computational Antitrust Project announces the publication of “Computational Presumptions Applied to AI Markets” by Alba Ribera Martínez. The article appears in Volume 6 of Stanford Computational Antitrust (pp. 32-67).

Digital regulators worldwide are imposing sweeping bans on data combinations to eliminate data asymmetries and learning effects. The paper argues that these interventions reveal a critical disconnect. Rules such as those introduced by the EU’s Digital Markets Act and the UK’s Digital Markets, Competition and Consumers Act were designed for traditional platforms. They are now being applied to AI downstream markets whose competitive dynamics differ. Reinforcement learning and model drift disrupt the standard feedback loops. The distinction between across-user and within-user learning further complicates the picture.

Building on this diagnosis, the article proposes an alternative instrument. Computational presumptions grounded in verifiable privacy-utility thresholds can serve as measurable compliance mechanisms. The paper identifies a NIST-aligned threshold of 2 < ε < 8 as a workable safe harbor, and shows how the same logic can extend to federated learning and homomorphic encryption through functionally equivalent indicators. The resulting framework transforms the regulator’s task from tracking every prohibited data combination to auditing a single measurable parameter.

Thibault Schrepel, founder of the Stanford Computational Antitrust Project, stated:

“Alba Ribera Martínez offers one of the clearest demonstrations that the DMA’s architecture does not transpose cleanly to AI markets. Her proposal gives regulators something measurable to audit instead of an unmanageable monitoring task. This is constructive legal scholarship.”

Alba Ribera Martínez is a Visiting Professor at the Brussels Study Centre and an Editor-in-Chief of Stanford Computational Antitrust. The article is available for download on the Stanford Computational Antitrust project’s page.