Moral Susceptibility and Robustness under Persona Role-Play in Large Language Models

Researchers analyzed how large language models express moral judgments when prompted to role-play different personas. The study found that Claude models are most morally robust, while larger models within families tend to be more susceptible to moral shifts through persona conditioning.

• →Claude AI models demonstrated the highest moral robustness among tested LLM families, followed by Gemini and GPT-4.
• →Larger language models within the same family show greater moral susceptibility to persona role-play prompts.
• →Model family accounts for most variance in moral robustness, while model size has no systematic effect on robustness.
• →The research introduces new benchmarks for measuring moral susceptibility and robustness in AI systems.
• →Moral robustness and susceptibility show positive correlation, particularly at the model family level.