AI Safety and Competition

Résumé

This paper examines how competition affects the timing of AI deployment under safety risk. We show that competition can generate two distortions relative to joint–profit maximization: a race to the bottom and insufficient entry. A race to the bottom arises when first-mover advantages induce premature deployment and is more likely as technological correlation (homogenization) increases. Conversely, firms may delay entry to free-ride on rivals’ experimentation, leading to insufficient entry. Even when private incentives under joint–profit maximization are aligned with social incentives, competition can still induce socially inefficient early deployment. We discuss policy implications for improving deployment timing.

Mots-clés

Codes JEL

Référence

Jay Pil Choi, Doh-Shin Jeon et Domenico Menicucci, « AI Safety and Competition », TSE Working Paper, n° 26-1745, mai 2026.