WHEN MACHINES FIX PRICES: THE ADEQUACY OF TRADITIONAL‘AGREEMENT’DOCTRINEIN ANTITRUSTLAW FOR REGULATING AI-DRIVEN ALGORITHMIC COLLUSION

Competition law has traditionally required proof of an agreement, tacit or explicit, between rival firms before conduct can be condemned as collusive. Self-learning pricing algorithms unsettle that foundation: independent machine-learning systems can converge on supra-competitive prices without any communication between competitors, raising the question of whether decades-old “meeting of the minds” doctrine can meaningfully police conduct that no human ever coordinated. This tension has moved from academic speculation into active litigation and legislation. The U.S. Department of Justice pursued RealPage over a rent-setting algorithm that aggregated non-public data from competing landlords, arguing that shared algorithmic infrastructure can facilitate coordination even absent an explicit price-fixing agreement; private multidistrict litigation has since produced settlements exceeding $359 million while related state-level suits continue into 2026. California has amended its Cartwright Act through Assembly Bill 325, effective January 1, 2026, to directly prohibit the use of “common pricing algorithms” built on shared competitor data
— a legislative acknowledgment that existing agreement-based doctrine may be too narrow. At the municipal level,
Berkeley’s ban on algorithmic rent-setting tools remains suspended pending resolution of a First Amendment
challenge brought by RealPage, illustrating how contested and unresolved this area still is even at the sub-state level. This paper asks whether statutory intervention of the California type — targeting the algorithmic mechanism itself rather than proving an underlying agreement — represents a durable doctrinal fix or merely a jurisdiction-specific patch that leaves federal Sherman Act enforcement structurally unable to reach purely algorithmic coordination. Drawing on the RealPage litigation record, the text of AB 325, and the comparative treatment of algorithmic pricing under Article 101 TFEU in EU competition law, the paper develops a typology of algorithmic coordination, situates each type against the doctrinal requirements of the Sherman Act, the amended Cartwright Act, and Article 101 TFEU, and argues that a durable framework requires re-examining what “agreement” should mean when the coordinating actor is not a human decision-maker but a selfoptimizing system trained on shared market data.