Only three AI models finished above starting capital in a 500-day startup survival test

Researchers at Princeton University developed a simulation called CEO-Bench to test the ability of AI agents to manage a fictional software company over 500 simulated days. Out of the models tested, only three managed to conclude the trial with more capital than they began with, while a basic rule-based program outperformed most AI systems. The results highlight significant limitations in the current capacity for autonomous agents to handle long-term business decision-making and resource management.