The Economic Behavior of AI Agents: Rationality and Reasoning in Large Language ModelsAs large language models (LLMs) take on increasingly consequential roles as autonomous decision-makers, understanding their economic reasoning and fairness is essential. We present a unified behavioral framework that combines experimental economics, game-theoretic analysis, and fairness assessment to rigorously evaluate how LLMs navigate uncertainty, interact strategically, and respond to demographic context. Our study demonstrates that these models consistently display bounded rationality, exhibiting systematic patterns in risk and loss sensitivity, and distinct reasoning depths across a spectrum of strategic settings. Critically, we show that LLMs’ decision processes are shaped not only by their underlying architectures but also by contextual factors such as demographic cues, revealing nuanced fairness concerns that go beyond surface-level outputs. By bridging behavioral science with computational evaluation, our approach establishes a robust foundation for diagnosing, comparing, and ultimately aligning the economic behavior of LLMs for real-world applications where both performance and ethical considerations are paramount.
