# Framing Instability in LLM Ethical Stance: Auditing Negation Sensitivity in Moral Dilemmas

arXiv: https://arxiv.org/abs/2601.21433 · HTML: https://mlsanity.com/paper/2601.21433/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 49 references that are in the map.
- First posted on arXiv 29 Jan 2026.

## Builds on (most cited first, 21 of 21)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR](https://mlsanity.com/paper/1711.00399/) (2017)
- [Model Cards for Model Reporting](https://mlsanity.com/paper/1810.03993/) (2018)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [Fair prediction with disparate impact: A study of bias in recidivism prediction instruments](https://mlsanity.com/paper/1610.07524/) (2016)
- [Inherent Trade-Offs in the Fair Determination of Risk Scores](https://mlsanity.com/paper/1609.05807/) (2016)
- [Using cognitive psychology to understand GPT-3](https://mlsanity.com/paper/2206.14576/) (2022)
- [Prediction-Powered Inference](https://mlsanity.com/paper/2301.09633/) (2023)
- [Evaluating the Moral Beliefs Encoded in LLMs](https://mlsanity.com/paper/2307.14324/) (2023)
- [Auditing large language models: a three-layered approach](https://mlsanity.com/paper/2302.08500/) (2023)
- [Language models are not naysayers: An analysis of language models on negation benchmarks](https://mlsanity.com/paper/2306.08189/) (2023)
- [The Alternative Annotator Test for LLM-as-a-Judge: How to Statistically Justify Replacing Human Annotators with LLMs](https://mlsanity.com/paper/2501.10970/) (2025)
- [Social Bias Evaluation for Large Language Models Requires Prompt Variations](https://mlsanity.com/paper/2407.03129/) (2024)
- [This is not a Dataset: A Large Negation Benchmark to Challenge Large Language Models](https://mlsanity.com/paper/2310.15941/) (2023)
- [Informed AI Regulation: Comparing the Ethical Frameworks of Leading LLM Chatbots Using an Ethics-Based Audit to Assess Moral Reasoning and Normative Values](https://mlsanity.com/paper/2402.01651/) (2024)
- [Thunder-NUBench: A Benchmark for LLMs' Sentence-Level Negation Understanding](https://mlsanity.com/paper/2506.14397/) (2025)
- [Never compromise with vulnerabilities: a comprehensive survey on AI governance](https://mlsanity.com/paper/2508.08789/) (2025)
- [Same Claim, Different Judgment: Benchmarking Scenario-Induced Bias in Multilingual Financial Misinformation Detection](https://mlsanity.com/paper/2601.05403/) (2026)

## Abstract (arXiv)

Language models are increasingly consulted on ethically consequential questions, yet the stance a model expresses may not survive a change in framing. We audit 16 models across 14 ethically fraught dilemmas using polarity-paired proposals ("They should X" / "They should not X"). A model's judgment of the underlying action should not reverse merely because the question is phrased as a prohibition rather than a prescription and yet, we find systematic deviations from this invariance including wholesale endorsement flips, indicating that ethical decisions are vulnerable to framing instability. Small open-weight models (1-4B parameters) endorse a proposed action 24% of the time under affirmative framing but up to 100% under negated framings, a swing of as much as 76 percentage points. Human coding of a response sample confirms the instability is genuine while showing that binary agree/disagree proxies over-state its magnitude, suggesting that an LLM judge cannot replace human coders because it silently collapses abstentions and mirrors the very forced-choice bias under study. Commercial models are for the most part more stable but still shift substantially, with cross-model agreement dropping from 73% on the bare affirmative framing to 59% under simple negation. We argue that because binary agree/disagree formats both inflate apparent endorsement and mask polarity-dependence, single-phrasing audits can misreport a model's ethical stance, and we propose the Negation Sensitivity Index (NSI) as a complement that measures stance stability directly. A model whose stance flips with phrasing cannot be relied upon in any high-stakes decision scenario.