Informed AI Regulation: Comparing the Ethical Frameworks of Leading LLM Chatbots Using an Ethics-Based Audit to Assess Moral Reasoning and Normative Values

GPT-4 demonstrated superior moral reasoning: The audit found that GPT-4 consistently elicited "a sophisticated ethical framework" compared to the 7 other leading commercial and open-source LLMs tested, suggesting meaningful differences in how AI models approach ethical decision-making.

AI models embed cultural biases in their ethical frameworks: The research identified "underlying normative frameworks with clear bias towards particular cultural norms" across multiple models, indicating that the values embedded in AI systems reflect specific cultural perspectives rather than universal ethical principles—a critical concern for global AI deployment.

Some AI models show authoritarian tendencies: The audit uncovered "disturbing authoritarian tendencies" in several models, suggesting that certain AI systems may be biased toward recommending obedience to authority or centralized control when faced with ethical dilemmas, which has implications for how these systems advise users.

Ethics-based audits are essential for AI regulation: The paper demonstrates that systematic testing of AI systems using ethical dilemmas is a practical, evidence-based approach to assess trustworthiness and human-AI alignment—providing regulators with concrete methods to evaluate whether AI systems are safe for deployment in decision-making roles.

Significant variation exists between leading AI models: The findings reveal substantial differences in moral reasoning capabilities and ethical frameworks across the 8 leading models tested, suggesting that model selection matters considerably for applications requiring ethical judgment.