Competition agencies worldwide now use AI tools routinely to detect cartels and bid-rigging, but the legal framework governing those tools has not kept pace

All 27 EU national competition agencies, the European Commission's DG Competition, and most competition agencies worldwide now rely on computational and artificial intelligence tools as standard enforcement instruments, according to analysis published by Vrije Universiteit Amsterdam academic Thibault Schrepel on 25 July 2026. The practice, known as 'computational antitrust', spans a wide range of applications: Spain screens public procurement for bid-rigging using BRAVA, a supervised machine learning system; Brazil processes procurement documents through a system called Cerebro; France queries its case database using a retrieval augmented generation (RAG) system built on large language models; Greece analyses email metadata seized in dawn raids; Chile monitors more than 80,000 products for price anomalies; and Poland investigates dark patterns using neuromarketing methods including eye-tracking. The analysis identifies two structural legal gaps. The first is a data collection problem: Regulation 1/2003 (the EU's primary competition enforcement regulation) and the ECN+ Directive (which extended investigative powers to national agencies) were drafted for document-request investigations, not the continuous large-scale machine-readable data flows that AI tools require. The second is a fairness problem: AI systems can produce biased outputs that no external party can examine, creating an explainability challenge that existing duty-to-give-reasons case law did not anticipate. The analysis also notes that the EU AI Act adds obligations in this space but does not settle how they apply to enforcement agencies using AI against private parties. A five-year research project called ATLANTIS, hosted at Vrije Universiteit Amsterdam and funded by an ERC Consolidator Grant, has been launched to address these doctrinal gaps.

Sign up to read →