Antitrust in the Age of AI

While the U.S., E.U., and U.K. can vary in their approaches to competition in tech, the three regulatory powerhouses have mirrored each other in their approaches to antitrust and competition in AI. South Korea, Japan, and India’s competition regulators, facing many of the same challenges, are similarly scrutinizing AI market dynamics amid antitrust concerns. 

Perhaps the most directly impactful and obvious challenge to consumers is the potential use of AI in pricing algorithms. In response to the DOJ and FTC’s recent intervention in RealPage and Yardi, and their action against Agri Stats concerning alleged AI-enabled hub-and-spoke agreements, industry will adapt quickly to avoid confidential information sharing and autonomous pricing. However, AI pricing programs powerful enough to independently develop coordination strategies could facilitate tacit collusion far more malevolent than traditional hub-and-spoke agreements: optimizing cartel gains, monitoring deviations instantaneously, and minimizing cheating. 

In May, a Federal Judge in Nevada dismissed the class action suit brought in Gibson, putting the judiciary at odds with the more robust interpretation advocated for by the DOJ and FTC in RealPage, Yardi, and Agri Stats. The court dismissed the claims after finding Las Vegas hotel operators did not share confidential information with the AI tool in question and were free to reject the tool’s pricing recommendations. If Gibson is a guide, then future enforcement would depend on whether such sharing of confidential information and lack of agency in pricing are required to prove collusion and whether independently collaborative AI systems are more akin to parallelism or hub-and-spoke agreements. 

Beyond consumer pricing, regulators continue to grapple with big tech’s predominance in hardware, computing, and capital, and the outsized influence that upstream market power grants over foundational technologies. As scrutiny of traditional M&A increased, a handful of tech giants have utilized an “interconnected web” of over 90 “partnerships” and “strategic investments” to assume control over the most advanced technologies. Regulators are investigating several of these investments and partnerships, such as those providing critical access to cloud computing. Such arrangements have thus far enjoyed early success in securing access to advanced foundation model technologies while avoiding the most scrutinizing antitrust enforcement.

Data is a similarly critical input in the AI value chain. Mass troves of data are essential in training generative pretrained transformers (GPT) models. Potential antitrust interest could take several directions. For example, Lina Khan, Chair of the FTC, recently stated that AI companies’ scraping of online content may constitute unfair competition. Whereas others are starting to bring claims alleging that AI companies have violated their IP rights by scraping content from the web. Such challenges have driven many AI companies towards favorable licensing deals, which the DOJ has expressed an interest in as an exercise of monopsony power. These enforcement efforts, however, could potentially harm the countervailing interest in greater AI model competition.

Similarly, proposals for restrictive regulation on AI foundation technologies, rather than end-use applications, have been criticized for picking winners and losers. Of concern is the disparate impact to entrants versus incumbents, namely around the potential for an increase in costs for new companies looking to enter and innovate. This is the most significant critique of the 2023 Executive Order on AI and the E.U. AI Act, which correlates AI model compute and a capacity to do harm. To be seen is whether this will lead to one or more licensing standards, such as a federal compulsory licensing regime in the U.S., such as is being proposed by some companies.