The Fragility Of Moral Judgment In Large Language Models

ArXi:2603.05651v1 Announce Type: cross People increasingly use large language models (LLMs) for everyday moral and interpersonal guidance, yet these systems cannot interrogate missing context and judge dilemmas as presented. We Surface perturbations produce low flip rates (7.5%), largely within the self-consistency noise floor (4-13%), whereas point-of-view shifts induce substantially higher instability (24.3%). A large subset of dilemmas (37.9%) is robust to surface noise yet flips under perspective changes, indicating that models condition on narrative voice as a pragmatic cue.