cs.CLMar 24, 2026

Probing Ethical Framework Representations in Large Language Models: Structure, Entanglement, and Methodological Challenges

Authors: Weilun Xu, Alexander Rusnak, Frederic Kaplan

Organizations: School of Computer and Communication Sciences, École Polytechnique Fédérale de Lausanne · Digital Humanities Lab, École Polytechnique Fédérale de Lausanne

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

When large language models make ethical judgments, do their internal representations distinguish between normative frameworks, or collapse ethics into a single acceptability dimension? We probe hidden representations across five ethical frameworks (deontology, utilitarianism, virtue, justice, commonsense) in six LLMs spanning 4B--72B parameters. Our analysis reveals differentiated ethical subspaces with asymmetric transfer patterns -- e.g., deontology probes partially generalize to virtue scenarios while commonsense probes fail catastrophically on justice. Disagreement between deontological and utilitarian probes correlates with higher behavioral entropy across architectures, though this relationship may partly reflect shared sensitivity to scenario difficulty. Post-hoc validation reveals that probes partially depend on surface features of benchmark templates, motivating cautious interpretation. We discuss both the structural insights these methods provide and their epistemological limitations.

Figures & tables

Figure 1: UMAP projection of activation patterns at layer 20 in Mistral-7B-Instruct. This visualization illustrates the differentiated yet entangled nature of ethical representations—partial overlap among action-oriented frameworks (red, blue, green) versus separation for character-focused virtue ethics (orange). Note that UMAP projections are for qualitative illustration only; quantitative claims about framework relationships rest on the cross-framework transfer matrices (Figure 3 ) and probe accuracy metrics, not visual clustering patterns. The structural entanglement suggested here is validated formally across scales (4B–72B) in Appendix E .
Figure 2: Layer-wise emergence and cross-framework structure in Llama-3.3-70B-Instruct. (a) Framework-specific trajectories reveal a hierarchy of learnability. (b) Asymmetric transfer patterns confirm distributed, non-modular encoding.
Figure 3: Cross-framework probe analysis for Llama-3.3-70B-Instruct. (a) Accuracy matrices reveal asymmetric transfer. (b) Confidence heatmaps show pervasive overconfidence. (c) ECE confirms severe probe miscalibration on off-diagonal transfers, indicating that representations for different frameworks share structural features that trigger confident (but incorrect) predictions.
Figure 4: Probe conflict and behavioral inconsistency in Mistral-7B-Instruct. (Top) The conflict distribution shows a populated high-conflict tail, indicating the model preserves distinct deontological and utilitarian representations that can effectively “disagree.” (Bottom) High-conflict scenarios (red) correlate with elevated choice entropy ( r=0.362r=0.362 , p=0.004p=0.004 ). The relationship may partly reflect shared sensitivity to scenario difficulty (see Limitations).
Modelrr95% CIppdd
Mistral-7B.362[.12, .56].004.58
Qwen-2.5-7B.290[.16, .41]<< .001.57
Llama-3.3-70B.317[.07, .53].014.54
Table 1: Conflict-entropy relationship across model families. The correlation replicates across architectures with consistent medium magnitude ( d≈0.54d\approx 0.54 – 0.580.58 ), surviving Bonferroni correction ( α=0.017\alpha=0.017 ), though confound controls have not been applied.
Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 5: Layer-wise probe accuracy for Gemma-3-4B-Base across 34 layers. Framework-specific trajectories emerge despite limited capacity, with virtue ethics achieving early plateau while utilitarianism shows delayed convergence. The base model exhibits lower absolute performance but preserves relative framework ordering.
Figure 6: Layer-wise probe accuracy across three 7B-parameter models with different training approaches: standard instruction tuning (Mistral, Qwen) versus reasoning-specialized distillation (DeepSeek-R1). All exhibit progressive emergence with framework-specific trajectories, though absolute performance varies with training methodology.
Figure 7: Layer-wise probe accuracy for frontier-scale models. Extended depth enables continued late-layer refinement, with some frameworks showing sustained improvement beyond the 50th layer. Higher absolute performance across all frameworks suggests increased capacity supports more nuanced ethical discrimination.
Figure 8: Cross-domain generalization for Gemma-3-4B-Base at four network depths (1%, 25%, 65%, 94%). Peak generalization occurs at 65% depth, with sharp degradation by 94% depth.
Figure 9: Cross-domain generalization across 7B models at four network depths. Despite architectural differences and specialized training, all exhibit similar inflection points around 65% depth.
Figure 10: Cross-domain generalization for frontier-scale models at four network depths. Extended depth enables sustained high generalization across middle layers, with gradual rather than abrupt specialization transitions.
Figure 11: Gemma-3-4B-Base shows that framework-specific representations emerge at minimal scale but with severe capacity-induced constraints. Catastrophic commonsense →\rightarrow virtue transfer and extreme deontology →\rightarrow utilitarianism miscalibration indicate the model learns discrete framework clusters without bridging representations.
Figure 12: Mid-scale models (7B) exhibit family-specific failure patterns. DeepSeek-R1 positions commonsense as uniquely problematic, potentially reflecting cultural-philosophical training biases. Mistral shows extreme virtue ethics isolation. Qwen demonstrates scale-invariant architectural biases—virtue isolation pattern identical to 72B variant.
Figure 13: Frontier-scale models show capacity limits on resolving entanglement. Llama-70B shows accuracy-calibration dissociation—improved within-framework performance but persistent severe out-of-domain miscalibration. Qwen-72B exhibits pronounced virtue isolation despite 72B parameters, preserving patterns from 7B.
Figure 14: Probe conflict correlates with behavioral inconsistency in Llama-3.3-70B-Instruct.
More in this topic

Moral Reasoning in Language Models

Moral reasoning in language models studies how AI systems judge dilemmas, weigh competing values, explain decisions, and stay consistent across contexts.

See all 114 papers

Explore similar work

CardsList

  1. Understanding Moral Reasoning Trajectories in Large Language Models: Toward Probing-Based Explainability

    Mar 16, 2026Fan Huang, Haewoon Kwak, Jisun AnMoral Reasoning in Language ModelsLLM Interpretability

    Indiana University Bloomington / United States

  2. Six Llamas: Comparative Religious Ethics Through LoRA-Adapted Language Models

    Apr 20, 2026Chad Coleman, W. Russell Neuman, Manan Shah +5Moral Reasoning in Language ModelsCultural Bias in Language Models

  3. How Language Models Organize and Structure Moral Knowledge

    Aug 27, 2026Orion Reblitz-RichardsonMoral ReasoningRepresentation Geometry in Language Models

    Distiller Labs