![[2602.13274] ProMoral-Bench: Evaluating Prompting Strategies for Moral Reasoning and Safety in LLMs](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13274] ProMoral-Bench: Evaluating Prompting Strategies for Moral Reasoning and Safety in LLMs
Summary
The paper introduces ProMoral-Bench, a benchmark for evaluating prompting strategies in large language models (LLMs) focused on moral reasoning and safety, revealing that simpler, exemplar-guided prompts outperform complex ones in performance and robustness.
Why It Matters
As AI systems increasingly influence decision-making, understanding how to effectively prompt LLMs for moral reasoning is crucial. ProMoral-Bench provides a standardized framework that can enhance the safety and ethical alignment of AI, addressing growing concerns about AI behavior in sensitive contexts.
Key Takeaways
- ProMoral-Bench evaluates 11 prompting strategies across four LLM families.
- Exemplar-guided prompts yield higher Unified Moral Safety Scores (UMSS) than complex multi-stage reasoning.
- The framework promotes cost-effective prompt engineering for better moral stability.
- Multi-turn reasoning is less robust under perturbations compared to few-shot exemplars.
- The study addresses fragmented empirical comparisons in LLM prompting strategies.
Computer Science > Artificial Intelligence arXiv:2602.13274 (cs) [Submitted on 5 Feb 2026] Title:ProMoral-Bench: Evaluating Prompting Strategies for Moral Reasoning and Safety in LLMs Authors:Rohan Subramanian Thomas, Shikhar Shiromani, Abdullah Chaudhry, Ruizhe Li, Vasu Sharma, Kevin Zhu, Sunishchal Dev View a PDF of the paper titled ProMoral-Bench: Evaluating Prompting Strategies for Moral Reasoning and Safety in LLMs, by Rohan Subramanian Thomas and 6 other authors View PDF HTML (experimental) Abstract:Prompt design significantly impacts the moral competence and safety alignment of large language models (LLMs), yet empirical comparisons remain fragmented across datasets and this http URL introduce ProMoral-Bench, a unified benchmark evaluating 11 prompting paradigms across four LLM families. Using ETHICS, Scruples, WildJailbreak, and our new robustness test, ETHICS-Contrast, we measure performance via our proposed Unified Moral Safety Score (UMSS), a metric balancing accuracy and safety. Our results show that compact, exemplar-guided scaffolds outperform complex multi-stage reasoning, providing higher UMSS scores and greater robustness at a lower token cost. While multi-turn reasoning proves fragile under perturbations, few-shot exemplars consistently enhance moral stability and jailbreak resistance. ProMoral-Bench establishes a standardized framework for principled, cost-effective prompt engineering. Subjects: Artificial Intelligence (cs.AI); Computation and Language (...
