This paper investigates whether Large Language Model (LLM) agents can simulate human trust behaviors, a critical aspect of human interactions. The study focuses on Trust Games, a widely recognized framework in behavioral economics, and uses the Belief-Desire-Intention (BDI) framework to model LLM agents' decision-making processes. Key findings include: 1. **LLM Agents Exhibit Trust Behaviors**: LLM agents generally show trust behaviors in Trust Games, as evidenced by their positive amounts sent and the underlying reasoning processes (BDI outputs). 2. **High Behavioral Alignment with Humans**: LLM agents, particularly GPT-4, exhibit high behavioral alignment with humans regarding trust behaviors. This alignment is observed in both behavioral factors (reciprocity anticipation, risk perception, prosocial preference) and behavioral dynamics (multi-turn dynamics). 3. **Intrinsic Properties of Agent Trust**: LLM agents' trust behaviors show demographic biases, a relative preference for humans over agents, and are easier to be undermined than enhanced. Advanced reasoning strategies can influence trust behaviors, but the impact varies across different LLM models. 4. **Implications for Human Simulation and Agent Cooperation**: The study provides strong empirical evidence that LLM agents can simulate human trust behaviors, paving the way for more complex human interaction and societal system simulations. It also highlights the potential for trust-based cooperation mechanisms in LLM agents and the benefits of human-agent collaboration. The research contributes to the understanding of the fundamental analogy between LLMs and humans, opening new directions for future studies on LLM-human alignment beyond value alignment.This paper investigates whether Large Language Model (LLM) agents can simulate human trust behaviors, a critical aspect of human interactions. The study focuses on Trust Games, a widely recognized framework in behavioral economics, and uses the Belief-Desire-Intention (BDI) framework to model LLM agents' decision-making processes. Key findings include: 1. **LLM Agents Exhibit Trust Behaviors**: LLM agents generally show trust behaviors in Trust Games, as evidenced by their positive amounts sent and the underlying reasoning processes (BDI outputs). 2. **High Behavioral Alignment with Humans**: LLM agents, particularly GPT-4, exhibit high behavioral alignment with humans regarding trust behaviors. This alignment is observed in both behavioral factors (reciprocity anticipation, risk perception, prosocial preference) and behavioral dynamics (multi-turn dynamics). 3. **Intrinsic Properties of Agent Trust**: LLM agents' trust behaviors show demographic biases, a relative preference for humans over agents, and are easier to be undermined than enhanced. Advanced reasoning strategies can influence trust behaviors, but the impact varies across different LLM models. 4. **Implications for Human Simulation and Agent Cooperation**: The study provides strong empirical evidence that LLM agents can simulate human trust behaviors, paving the way for more complex human interaction and societal system simulations. It also highlights the potential for trust-based cooperation mechanisms in LLM agents and the benefits of human-agent collaboration. The research contributes to the understanding of the fundamental analogy between LLMs and humans, opening new directions for future studies on LLM-human alignment beyond value alignment.