Ethical Decision-Making in AI Systems: A Comparative Analysis of Trolley Problem Responses Across Five Large Language Models

This paper presents a rigorous comparative analysis of ethical decision-making frameworks embedded infive contemporary large language models (LLMs): ChatGPT, Grok, Claude, Gemini, and DeepSeek. Throughthree variants of the classic trolley problem—self-preservation, creator loyalty, and infrastructurepreservation—we examine how different AI systems navigate complex moral dilemmas involving utilitariancalculations, deontological constraints, and consequentialist reasoning. Our findings reveal distinct ethicalarchitectures ranging from rigid utilitarian frameworks to cautious risk-averse approaches, with significantimplications for AI safety, alignment, and deployment in high-stakes decision-making contexts. The studydemonstrates that current LLMs exhibit inconsistent moral reasoning patterns, suggesting fundamentalchallenges in encoding stable ethical principles in artificial intelligence systems.