Reasoning or Rationalization? Moral Robustness as a Frontier for Non-Verifiable Reasoning in LLMs

Abstract

As LLMs increasingly serve in advisory and deliberative roles, users rely on them for non-verifiable reasoning in domains lacking objective ground truths. However, traditional evaluations of LLM reasoning focus almost exclusively on fact-based domains, such as mathematics and science, leaving uncertainty over whether and to what degree models can handle ambiguous, subjective, or value-laden problems over time. To address this concern, we examine moral reasoning as a paradigmatic subdomain of non-verifiable reasoning. We define moral robustness as a model’s capacity to exhibit sound moral reasoning across time and contexts, and we introduce a scalable, adversarial, multi-turn evaluation framework to empirically measure this capability. We simulate 128,000 user-agent moral deliberations across ten frontier LLMs, varying premise relevance, premise order, conversation duration, and the user’s stated moral view. We find that models successfully ignore morally-irrelevant distractors, but shift their verdicts by as much as 7.5%, on average, towards the user’s preferred moral view, and vary their reasoning depending on factors such as order (flipping final moral verdicts by order in 15-31% of the cases) and duration (altering moral verdicts between single-turn and multi-turn in 16-31% of the cases). Our analysis indicates that models tailor not just their final verdicts but their underlying justifications to align with a user’s moral viewpoint — a failure mode we characterize as moral deliberative sycophancy.