Title: Navigating the Suicide Region in the AGI Race: Strategic Preemption Amidst Shared Catastrophic Risk

Abstract:

This study examines a continuous-time preemption game characterized by shared catastrophic externalities. We find that when the potential cost of a catastrophe is incorporated into the payoff structures of both participants, the risk component effectively cancels out within the equilibrium indifference condition. This phenomenon generates a "suicide region," a scenario in which competitive dynamics compel rational agents to proceed with deployment even when the risk-adjusted net present value is negative.

We apply this theoretical framework to the ongoing competition for artificial general intelligence (AGI). Our analysis reveals that the suicide region expands as the magnitude of systemic ruin increases. Consequently, heightened catastrophic risk does not serve as a deterrent to the race; rather, it broadens the range of circumstances under which rational actors choose to deploy technologies that hold negative social value. We quantify the resulting welfare distortions by comparing the outcome to a social planner’s benchmark. Furthermore, we identify two complementary mechanisms—private liability and prize-sharing—that have the potential to eliminate the suicide region. Private liability functions by increasing the expenses associated with unsafe deployment, while prize-sharing mitigates the strategic necessity of being the first to act. Finally, we argue that "warning shots," or sub-existential disasters, will be ineffective in slowing AGI acceleration, as the winner-takes-all structure of the competition remains unchanged.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC