A new working paper from the University of Chicago’s Harris School of Public Policy argues that intensifying competition among AI developers may be making the race to artificial general intelligence more dangerous, not less, by creating economic incentives to prioritise speed over safety.
The research, authored by Ethan Bueno de Mesquita, dean and Sydney Stein Professor at Harris, Wioletta Dziuda, associate professor and deputy dean for faculty and research, and Mattias Polborn of Vanderbilt University, models how firms competing to be first to achieve AGI allocate resources between capability development and safety. Their central finding is that as the number of competing firms increases, each devotes a greater share of resources to speed and a smaller share to safety, raising the probability of harmful outcomes across the industry.
The researchers describe the dynamic as a collective-action problem. Individual firms may privately prefer a slower, safer race, but competitive pressure makes unilateral caution economically irrational. The model also finds that firms may continue racing even when the expected value of achieving AGI turns negative, because the catastrophic downside affects all players regardless of whether they participate.
The paper challenges conventional assumptions that market competition reliably produces better outcomes, and offers a framework for evaluating AI governance policies. The authors find that restricting computing resources is not always beneficial, and that in some market configurations, providing additional resources alongside limits on the number of competitors may better support safety. They also identify merit in publicly funded AI development prioritising safety over speed, citing Switzerland as an emerging example of this model.
