A Scalable Approach to Evaluating Moral Sensitivity in LLMs

2026-07-03

AI Summary

This work proposes a scalable evaluation framework for assessing the stability of large language models (LLMs) in identifying morally relevant attributes, without relying on human annotations or the models’ own moral judgment capabilities. The approach procedurally generates moral scenarios with controlled perturbations to construct the MORPH-1K benchmark, and evaluates output invariance through textual noise injection and semantic similarity analysis. Experiments across eight mainstream LLMs demonstrate that, although perturbations affect the number of identified moral attributes, the semantic content of these attributes remains highly stable—exhibiting similarity scores significantly above threshold levels. These findings validate the framework’s effectiveness and scalability in evaluating moral sensitivity in LLMs.

Abstract

Moral sensitivity is the ability to identify the morally relevant features of a decision situation and use them as the basis for action. It is the foundation of broader moral competence: any other moral reasoning capabilities will be irrelevant if an agent lacks sensitivity to the relevant facts. In this paper, we offer a new evaluation of LLM moral sensitivity and in doing so, we address and resolve a central problem in AI alignment research: how to scale behavioural evaluations beyond expensive and sometimes metaethically dubious comparisons with a human baseline, without adopting an LLM judge that must be assumed to have the very capability that you are attempting to evaluate. Our central question is this: can LLMs successfully identify the morally relevant features of noisy cases, in which various kinds of morally irrelevant information have been introduced to distract the respondent? To explore this, we introduce \textbf{MORPH-1K (MOral Robustness under Perturbed Hypotheticals)}, a procedurally-generated 1,000-case benchmark spanning 50 moral foundation-pole combinations across four social domains. MORPH-1K is paired with a suite of textual noise elements, along with a method for validating that the distractors do not change the morally salient content of the case. We apply MORPH-1K to eight contemporary LLMs, and show that while morally irrelevant perturbations often changed the number of features listed, the semantic content of those features remained stable across all noise conditions, with similarity scores above our calibrated floor threshold. More broadly, our invariance framework extends to evaluative domains where ground truth is difficult to specify but relevant and irrelevant features can be separated by design.

Problem

• moral sensitivity
• LLM evaluation
• AI alignment
• behavioral evaluation
• scalable benchmark

Innovation

• moral sensitivity
• scalable evaluation
• procedurally-generated benchmark
• textual perturbations
• invariance framework