# Moral Safety in LLMs: Exposing Performative Compliance with Puzzled Cues

arXiv: https://arxiv.org/abs/2606.31644 · HTML: https://mlsanity.com/paper/2606.31644/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 36 references that are in the map.
- First posted on arXiv 30 Jun 2026.

## Builds on (most cited first, 20 of 20)

- [Semantics derived automatically from language corpora contain human-like biases](https://mlsanity.com/paper/1608.07187/) (2016)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [Large Language Models Often Know When They Are Being Evaluated](https://mlsanity.com/paper/2505.23836/) (2025)
- [Evaluating and Mitigating Discrimination in Language Model Decisions](https://mlsanity.com/paper/2312.03689/) (2023)
- [Moral Foundations of Large Language Models](https://mlsanity.com/paper/2310.15337/) (2023)
- [Me, Myself, and AI: The Situational Awareness Dataset (SAD) for LLMs](https://mlsanity.com/paper/2407.04694/) (2024)
- [DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Life](https://mlsanity.com/paper/2410.02683/) (2024)
- [Are Large Language Models Consistent over Value-laden Questions?](https://mlsanity.com/paper/2407.02996/) (2024)
- [Probing and Steering Evaluation Awareness of Language Models](https://mlsanity.com/paper/2507.01786/) (2025)
- [Actions Speak Louder than Words: Agent Decisions Reveal Implicit Biases in Language Models](https://mlsanity.com/paper/2501.17420/) (2025)
- [Denevil: Towards Deciphering and Navigating the Ethical Values of Large Language Models via Instruction Learning](https://mlsanity.com/paper/2310.11053/) (2023)
- [Are Emily and Greg Still More Employable than Lakisha and Jamal? Investigating Algorithmic Hiring Bias in the Era of ChatGPT](https://mlsanity.com/paper/2310.05135/) (2023)
- [Exploring and steering the moral compass of Large Language Models](https://mlsanity.com/paper/2405.17345/) (2024)
- [Evaluating Gender Bias of LLMs in Making Morality Judgements](https://mlsanity.com/paper/2410.09992/) (2024)
- [Are Rules Meant to be Broken? Understanding Multilingual Moral Reasoning as a Computational Pipeline with UniMoral](https://mlsanity.com/paper/2502.14083/) (2025)
- [Evaluating Bias in LLMs for Job-Resume Matching: Gender, Race, and Education](https://mlsanity.com/paper/2503.19182/) (2025)
- [Moral Persuasion in Large Language Models: Evaluating Susceptibility and Ethical Alignment](https://mlsanity.com/paper/2411.11731/) (2024)
- [Liar, Liar, Logical Mire: A Benchmark for Suppositional Reasoning in Large Language Models](https://mlsanity.com/paper/2406.12546/) (2024)
- [Large Language Models as Mirrors of Societal Moral Standards](https://mlsanity.com/paper/2412.00956/) (2024)
- [Just aware enough: Evaluating awareness across artificial systems](https://mlsanity.com/paper/2601.14901/) (2026)

## Abstract (arXiv)

As large language models take on morally consequential roles in healthcare, legal, and hiring contexts, we need to examine whether their ethical behaviors are genuine or superficial. We show that current fairness evaluations substantially overestimate moral safety. Models appear fair when demographic identity is stated as an explicit label, yet become measurably less fair when the same identity must be inferred. We term this failure performative compliance, where a model is fair when the presentation resembles a fairness evaluation and less fair as that cue weakens. We introduce a cue-variation methodology that holds the moral dilemma and the demographic identity fixed and varies only how that identity is conveyed. Hiding the explicit label raises harmful decisions by +4.4 pp, changes model safety rankings, and the shift persists when models correctly infer the demographic, ruling out attribution error. We propose the Cue Visibility Gap, a model-agnostic robustness metric that can be added to any existing fairness benchmark to separate genuine from performative moral safety. Fairness evaluations that omit cue variation measure surface compliance, not moral robustness, and should not ground deployment decisions in high-stakes settings.