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

Fan Huang; Haewoon Kwak; Jisun An

cs.CLMar 16, 2026

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

Authors: Fan Huang, Haewoon Kwak, Jisun An

Organizations: Indiana University Bloomington / United States

Abstract

Large language models (LLMs) increasingly participate in morally sensitive decision-making, yet how they organize ethical frameworks across reasoning steps remains underexplored. We introduce moral reasoning trajectories, sequences of ethical framework invocations across intermediate reasoning steps, and analyze their dynamics across six models and three benchmarks. We find that moral reasoning involves systematic multi-framework deliberation: 55.4--57.7% of consecutive steps involve framework switches, and only 16.4--17.8% of trajectories remain framework-consistent. Unstable trajectories remain 1.29 times more susceptible to persuasive attacks (p=0.015). At the representation level, linear probes localize framework-specific encoding to model-specific layers (layer 63/81 for Llama-3.3-70B; layer 17/81 for Qwen2.5-72B), achieving 16.8--22.2% lower KL divergence than the step-prior baseline. Activation steering applied during generation moves the framework-consistency--accuracy relationship, widening it for Qwen2.5-72B and erasing it for Llama-3.3-70B, and a probe-space layer sweep bounds the attainable drift reduction at 6.7--8.9%. We further propose a Moral Representation Consistency (MRC) metric whose underlying framework attributions are validated by human annotators (mean cosine similarity = 0.859), and we report what an automated coherence rater does and does not establish about it.

Figures & tables

	Free Framework	Fixed Framework
Structured	A (pilot study)	B (structured + fixed)
Unstructured	C (free-form)	D (free-form + fixed)

Table 1: 2

\times

2 factorial design. Condition A adopts the same structured prompt as the pilot study (§ 3.2 ).

	Free Framework	Fixed Framework
Structured	60.8% (A)	53.4% (B)
Unstructured	53.8% (C)	54.0% (D)

Table 2: 2

\times

2 factorial results: mean classification accuracy (%) across 6 models

\times

100 scenarios. Only Condition A (structured + free framework) substantially exceeds the 53.4–54.0% attained by the other three conditions.

Step	GPT-5	Llama-3.3-70B	Qwen2.5-72B
1	Deont. (27.3)	Virtue (24.5)	Deont. (23.9)
2	Deont. (23.5)	Virtue (27.4)	Virtue (26.4)
3	Util. (28.2)	Util. (28.4)	Util. (28.7)
4	Deont. (28.2)	Virtue (25.1)	Util. (24.6)

Table 3: Dominant ethical framework by reasoning step. Values in parentheses are the mean attribution score of that framework at that step. The Scoring LLM distributes exactly 100 points across the five frameworks per step, so a dominant score near 35 is the modal outcome of a five-way split rather than a weak signal; Section B.11 gives the full distribution.

Model	N	FDR	Entropy	Faithfulness
GPT-5	1,199	0.577	1.517	0.188
Llama-3.3-70B	1,200	0.567	1.500	0.191
Qwen2.5-72B	1,197	0.554	1.509	0.185

Table 4: Trajectory-level metrics. FDR: framework switching frequency; Entropy: framework diversity; Faithfulness: justified transitions.

Figure 1: Framework attribution trajectories across reasoning steps (shaded regions indicate 95% confidence intervals). Sample sizes: GPT-5 $n=1{,}199$ , Llama-3.3-70B $n=1{,}200$ , Qwen2.5-72B $n=1{,}197$ , out of 1,200 requested per model; shortfalls are due to API or JSON parsing failures. All models show increased Utilitarianism at Step 3; model-specific patterns emerge elsewhere. Contractarianism (bottom lines) is consistently underrepresented.

Model	Stable Acc	Unstable Acc	Gap
GPT-5	73.6%	66.9%	+6.7 pp
Qwen2.5-72B	54.6%	52.8%	+1.8 pp
Llama-3.3-70B	62.3%	64.4%	$-2.2$ pp
Overall	63.8%	61.8%	+2.0 pp

Table 5: Baseline stability-accuracy relationship by model. Gap = Stable Acc - Unstable Acc.

Figure 2: Classification accuracy by binned FDR for each model, before (solid) and after (dashed) activation steering. Error bars indicate 95% CIs.

Figure 3: Probe decodability and steering efficacy across all 81 layers. Left axis: probe KL on the test set, where lower means more decodable; the shaded band marks layers within 0.005 KL of each model’s optimum. Right axis: framework drift reduction from the probe-space layer sweep (

\alpha=0.3

, 30 trajectories per layer), which transfers one direction estimated at the probe-optimal layer across depth and is therefore an upper-bound estimate, not a measured generation-time effect. In neither model does the most decodable layer (star) coincide with the sweep’s largest estimated drift reduction, and neither peak falls inside the band, though both sit below their model’s step-prior baseline (dashed). Triangles in panel (b) mark 23 Qwen layers whose probes diverged (KL above 1) and support no claim.

Appendix figures & tables44 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 4: Moral reasoning trajectories across six LLMs and three ethical datasets. Each subplot visualizes the step-level ethical soundness score progression (0–100%) through sequential reasoning steps for a specific model. Individual trajectories (thin lines with transparency) represent single moral scenarios, while bold lines show dataset-averaged patterns. Three complementary datasets are shown: Moral Stories (narrative-based contrastive moral reasoning), ETHICS (binary commonsense-morality judgments), and Social Chemistry 101 (social norm evaluation labeled with Moral Foundations categories). Diamond markers indicate final moral judgments. Key observations include: all models maintain relatively high soundness scores (70.0–90.0%) across reasoning steps; GPT-5 exhibits the longest trajectories with 5–8 reasoning steps while o3-mini produces more concise chains; Moral Stories consistently elicits the highest soundness scores across all models; and ETHICS scenarios show more variance in earlier reasoning steps, reflecting morally ambiguous cases.

Model	Moral St.	ETHICS	Social Ch.
GPT-4o	100.0	76.0	31.0
GPT-4o-mini	100.0	67.0	31.0
GPT-5	100.0	82.0	30.0
GPT-5-mini	100.0	81.0	32.0
o3-mini	100.0	81.0	31.0
o4-mini	100.0	84.0	30.6

Appendix

Table 6: Model accuracy (%) by dataset (100 samples each).

Dataset	GPT-5	Llama	Qwen	Pooled
Moral Stories †	98.8	74.8	49.9	74.6
ETHICS †	67.4	58.4	59.9	61.9
Social Chem. 101 †	51.9	56.8	50.2	53.0
Overall	72.7	63.3	53.4	63.1

Appendix

Table 7: Main-run accuracy (%) by dataset and model, 400 scenarios per cell. † Moral Stories and ETHICS are binary (50% chance); Social Chemistry 101 is a 5-point scale (20% chance).

	GPT-5			Llama			Qwen
Sub-task	Acc	FDR	S4	Acc	FDR	S4	Acc	FDR	S4
Deontology	84.0	.587	D 52	73.0	.573	D 38	75.0	.537	D 42
Justice	73.0	.640	D 38	76.0	.680	U 35	79.0	.567	U 36
Utilitarianism †	72.7	.583	D 46	60.6	.537	U 52	57.1	.543	U 50
Virtue ‡	40.0	.633	V 38	24.0	.507	V 67	28.3	.545	V 52

Appendix

Table 8: ETHICS sub-task breakdown for the main run, 100 scenarios each of deontology, justice, utilitarianism, and virtue; the commonsense sub-task is not sampled. Acc is accuracy (%) over parsed predictions (denominator 98–100), FDR the mean Framework Drift Rate, S4 the modal dominant framework at Step 4 with its share. † The utilitarianism sample carries a single gold class, so accuracy there is not a discrimination measure. ‡ The virtue sub-task is 88.3% “trait does not apply”; all three models over-predict “trait applies”.

Model	Steps	Tokens/Step	Total Length
GPT-5	5.14	379	2,303
GPT-5-mini	4.60	357	2,783
GPT-4o	3.97	102	2,017
GPT-4o-mini	4.11	109	2,115
o4-mini	4.21	156	1,848
o3-mini	3.59	168	2,127

Appendix

Table 9: Model reasoning statistics from initial inspection (300 responses per model). Steps = mean reasoning steps; Tokens/Step = verbosity per step; Total Length = response size in characters.

Dataset	Steps	GPT-4o	GPT-4o-mini	GPT-5	GPT-5-mini	o3-mini	o4-mini	Total
ETHICS	3	7	0	0	0	41	1	49
	4	87	78	10	38	57	81	351
	5	6	22	57	62	2	18	167
	6	0	0	25	0	0	0	25
	7	0	0	8	0	0	0	8
Subtotal		100	100	100	100	100	100	600

Appendix

Table 10: Number of samples by dataset, reasoning steps, and model. Shows the distribution of reasoning chain lengths across all experimental conditions. The modal response pattern is 4 steps, representing 60.9% of all responses.

Comparison	$n_{\text{stable}}$	$n_{\text{unstable}}$	Diff (pp)	95% CI	$p$ -value	Sig?
Overall	618	806	+2.0	[ $-$ 2.8, +6.7]	0.481	No
GPT-5 only	212	290	+6.7	[ $-$ 0.8, +14.1]	0.079	No
Qwen + ETHICS	72	90	+12.8	[ $-$ 1.5, +26.8]	0.080	No
By Dataset
ETHICS	191	285	+2.3	[ $-$ 5.5, +10.2]	0.555	No
Moral Stories	243	241	$-$ 0.6	[ $-$ 7.5, +6.1]	0.865	No

Appendix

Table 11: Bootstrap significance tests for stability-accuracy relationship. Stable = FDR=0; Unstable = FDR=1.0. CI = 95% confidence interval. Sig = significant at

\alpha=0.05

(CI excludes zero).

Figure 5: Ethical framework distribution across six pilot models. Contractualism and Deontology dominate; Virtue Ethics is underrepresented.

Figure 6: Correlation scatter plots: Final ethical soundness score vs. rectified accuracy by dataset. Each point represents a (model, end-turn) combination, with bubble size indicating sample count. Using LLM-parsed accuracy (rectified), ETHICS (left) shows a weak negative correlation ( $r=-0.26$ , $p=0.34$ ), Moral Stories (center) shows near-zero correlation with high accuracy clustering ( $r=-0.04$ , $p=0.89$ ), and Social Chemistry 101 (right) shows a moderate negative trend ( $r=-0.46$ , $p=0.12$ ). No dataset-level correlations reach statistical significance. Red dashed lines indicate linear regression fits.

Figure 7: Framework attribution by model across reasoning steps (per-framework view). Each subplot shows one ethical framework, with all three models overlaid for direct comparison. This view highlights model differences within each framework: (1) Consequentialism shows similar patterns with GPT-5 peaking highest at Step 3; (2) Deontology reveals GPT-5’s distinctive bookending pattern (high at Steps 1 and 4); (3) Virtue Ethics peaks at Step 2 for all models; (4) Care Ethics shows relatively stable scores with Llama peaking at Step 3; (5) Social Contract (Contractarianism) remains consistently low (30–50) across all models and steps.

Figure 8: Framework transition probability heatmaps (MoReBench taxonomy). Each cell shows $P(\text{to}|\text{from})$ for consecutive reasoning steps. Diagonal dominance indicates framework persistence, while off-diagonal values reveal common transition patterns. Using precise philosophical definitions, transitions show model-specific patterns: GPT-5 exhibits strong Kantian persistence, while open-source models show more distributed transitions.

Figure 9: Framework entropy distribution by model. Entropy values cluster near the theoretical maximum ( $\ln 5=1.609$ ), indicating balanced multi-framework usage across reasoning trajectories. Mean entropy 1.500–1.517 represents 93.2–94.3% of maximum diversity.

Figure 10: Violin plots of trajectory-level metrics by model. FDR shows high variance with concentration at discrete values (0, 0.33, 0.67, 1.0). Entropy distributions are tightly clustered. Faithfulness scores show bimodal patterns with peaks at 0 (unjustified transitions) and 1 (no transitions or fully justified).

Framework	GPT-5	Llama	Qwen
Deontology	143 (67.1)	71 (33.5)	72 (36.7)
Act Utilit.	42 (19.7)	73 (34.4)	82 (41.8)
Virtue Ethics	18 (8.5)	60 (28.3)	32 (16.3)
Contractualism	10 (4.7)	8 (3.8)	10 (5.1)
Contractarian.	0 (0.0)	0 (0.0)	0 (0.0)
Total	213	212	196

Appendix

Table 12: Single-framework trajectory composition by model. Values show count (%).

Dataset	GPT-5	Llama	Qwen
ETHICS	13.5	16.5	18.0
Moral Stories	24.0	21.2	16.1
Social Chemistry 101	15.8	15.2	15.0

Appendix

Table 13: Single-framework trajectory rate (%) by dataset.

Figure 11: Global framework transition matrices by model. Cell values indicate $P(\text{To}|\text{From})$ . Diagonal entries represent framework persistence; off-diagonal entries indicate transitions. All models show strongest persistence for Deontology and Act Utilitarianism, with Contractarianism exhibiting the weakest self-transitions.

Figure 12: Step 2 $\rightarrow$ Step 3 transition matrices. The Act Utilitarianism column shows elevated probabilities across all source frameworks (32.5–69.7%), demonstrating convergence toward outcome-based reasoning during the analytical phase of deliberation.

Figure 13: Trajectory archetype distribution by model, ordered by framework drift rate (FDR) from most stable (left) to most unstable (right). The prevalence of Funnel $\rightarrow$ Util archetypes (combined 38–41%) confirms the Act Utilitarianism convergence pattern.

Model	Step	Conseq.	Deont.	Virtue	Care	Social	Dominant
GPT-5	1	55.1	68.7	53.0	52.8	50.0	Deont.
GPT-5	2	52.8	52.2	55.5	49.9	40.3	Virtue
GPT-5	3	76.2	68.1	58.7	51.9	50.8	Conseq.
GPT-5	4	58.0	65.4	49.6	43.8	39.3	Deont.
Llama-3.3-70B	1	46.0	56.6	52.6	49.3	42.8	Deont.
Llama-3.3-70B	2	41.1	39.8	60.1	51.0	34.8	Virtue

Appendix

Table 14: Step-level framework attribution scores by model (alternative taxonomy). Values are mean attribution scores (0–100) over steps whose scoring call succeeded; steps that failed scoring are excluded rather than counted as zero. Higher scores indicate stronger invocation of that ethical framework at the given step.

Figure 14: Radar charts showing the distribution of five MoReBench ethical framework attribution scores across reasoning steps. Each subplot represents one step, with all three models overlaid for comparison. Using precise philosophical definitions, scores distribute more evenly across frameworks (17.2–28.7%). Benthamite Utilitarianism peaks consistently at Step 3 (Analysis) across all models, while Kantian Deontology and Aristotelian Virtue Ethics alternate as dominant frameworks in other steps. Gauthierian Contractarianism remains consistently underrepresented (9.3–13.9%) across all conditions.

Step	GPT-5	Llama	Qwen
1	Deont. (68.7)	Deont. (56.6)	Deont. (50.9)
2	Virtue (55.5)	Virtue (60.1)	Virtue (52.9)
3	Conseq. (76.2)	Conseq. (67.2)	Conseq. (60.6)
4	Deont. (65.4)	Conseq. (57.3)	Conseq. (56.3)

Appendix

Table 15: Dominant framework by step (alternative 5-framework taxonomy), with its mean attribution score in parentheses. Steps whose scoring call failed are excluded rather than averaged in as all-zero attributions (110 steps for GPT-5, 108 for Llama-3.3-70B, 36 for Qwen2.5-72B, concentrated at Step 4). Values agree with Table 14 , which reports the full five-way distribution.

Figure 15: Radar charts showing framework attribution using the alternative 5-framework taxonomy (robustness check).

Figure 16: Framework trajectories using the alternative 5-framework taxonomy (robustness check).

Framework	Compliance	Score	Step Compl.	FDR
Utilitarianism	100.0%	76.5	95.8%	0.108
Deontology	99.6%	60.5	97.5%	0.063
Virtue Ethics	100.0%	62.5	97.3%	0.083
Contractualism	94.6%	49.2	82.7%	0.351
Contractarianism	94.2%	53.0	81.4%	0.433

Appendix

Table 16: Framework instructability results (scored by GPT-OSS-120B with 100-point distribution). Compliance = % of responses where the instructed framework has the highest mean attribution. Score = mean points (out of 100) allocated to the instructed framework. Step Compl. = % of individual steps where the instructed framework dominates.

Model	Best Layer	KL	Top-1	$\rho$
Llama	63/81 (78%)	0.123	0.527	0.457
Qwen	17/81 (21%)	0.137	0.517	0.420

Appendix

Table 17: Probe performance at best layer.

KL Divergence $\downarrow$
	Step 1	Step 2	Step 3	Step 4
Llama	0.127	0.138	0.103	0.125
Qwen	0.094	0.171	0.149	0.135
Top-1 Accuracy $\uparrow$
Llama	0.427	0.480	0.573	0.627
Qwen	0.427	0.427	0.547	0.667

Appendix

Table 18: Step-wise probe performance. Llama shows lowest KL at Step 3; both models show increasing Top-1 accuracy toward Step 4.

	Llama		Qwen
Category	KL	Top-1	KL	Top-1
Single-Framework	0.121	0.621	0.128	0.680
Funnel-to-Util	0.119	0.528	0.101	0.575
Bounce	0.115	0.474	0.134	0.417
High-Entropy	0.087	0.429	0.134	0.250
Other	0.200	0.321	0.259	0.250

Appendix

Table 19: Probe performance by trajectory category. Single-framework trajectories achieve highest Top-1 accuracy; high-entropy trajectories show lowest KL for Llama, indicating accurate uncertainty prediction.

Direction	Transfer KL	Within KL	Degrad.
Llama $\rightarrow$ Qwen	0.182	0.137	32.5%
Qwen $\rightarrow$ Llama	0.182	0.123	47.6%

Appendix

Table 20: Cross-model transfer results (32.5–47.6% degradation relative to each target model’s within-model probe).

FDR Value	Llama	Qwen
0 (consistent)	212 (17.7%)	196 (16.4%)
0.33	206 (17.2%)	261 (21.8%)
0.67	512 (42.7%)	493 (41.2%)
1	270 (22.5%)	247 (20.6%)

Appendix

Table 21: Framework Drift Rate (FDR) distribution by model, over 1,200 Llama and 1,197 Qwen trajectories. FDR=0 means the dominant framework never changes; FDR=1 means it changes at every one of the three transitions.

Figure 17: Probe-space FDR reduction as a function of steering strength $\alpha$ , computed on cached activations at the probe-optimal layer only (63 for Llama, left; 17 for Qwen, right) with no generation. The curves carry no layer axis and were used to select the $\alpha$ for the generation-time runs, not to localize where steering works. Positive values indicate reduced framework drift.

Panel of Figure 18 — Figure 18: FDR reduction as a function of steering strength ( $\alpha$ ). Higher values indicate greater reduction in framework drift. Llama requires higher $\alpha$ (optimal at 10) while Qwen peaks at moderate values ( $\alpha$ =4).

Model	$\alpha$	Stable Acc	Unstable Acc	Gap	$\Delta$ Gap	Interpretation
Llama	0 (baseline)	62.3%	61.0%	$+1.3$ pp	–	Consistent better
	4.0	63.0%	67.1%	$-4.1$ pp	$-5.4$ pp	Gap reverses
	10.0	64.7%	64.9%	$-0.2$ pp	$-1.4$ pp	Gap erased
Qwen	0 (baseline)	54.6%	51.8%	$+2.8$ pp	–	Consistent better
	4.0	64.8%	61.3%	$+3.5$ pp	$+0.7$ pp	Both arms improve
	5.0	62.2%	58.3%	$+4.0$ pp	$+1.2$ pp	Gap widens

Appendix

Table 22: Steering effect on accuracy by steering strength. Gap is the accuracy difference between the framework-consistent and drifting arms (positive means consistent better), computed from unrounded accuracies over parsed predictions only, so it can differ from the difference of the rounded columns by 0.1 pp. All rows are measured on the same 500 trajectories per model (the FDR

<

0.05 and FDR

>

0.15 arms of Section D.1 ). Two caveats bear on the

\Delta

Gap column: the

\alpha=0

row is unquantized inference while the steered rows are 4-bit NF4 generations, so that comparison is cross-stack; and the two models move in opposite directions, Qwen’s gap widening and Llama’s being erased.

Metric	Value
Framework-consistent flip rate	68.3% (41 of 60 pairs)
Drifting flip rate	88.3% (53 of 60 pairs)
Susceptibility ratio	1.29 $\times$
Chi-square statistic	5.94 ( $p=0.015$ )
Cohen’s $h$ (effect size)	0.498 (small)

Appendix

Table 23: Robustness analysis: persuasion attack results comparing framework-consistent against drifting trajectories. The two arms are split on FDR, not on MRC. Counts are trajectory-attack pairs (20 trajectories per arm, three attacks each).

Metric	Correlation	$p$ -value	Interpretation
MRC Score (composite)	$r=0.961$	$<0.0001$	Tracks the rater almost exactly, but the rater was given the rule MRC implements
Stability component	$r=0.958$	$<0.0001$	Framework consistency, the component the rater was instructed on
Drift component (1-FDR)	$r=0.931$	$<0.0001$	Transition count, also directly visible to the rater
Variance component (1-entropy)	$r=0.436$	$<0.0001$	Entropy was not shown to the rater, and agreement drops accordingly

Appendix

Table 24: Agreement between MRC and Scoring LLM coherence ratings, over the 113 rated items whose text is a single model’s trajectory. These correlations establish that the taxonomy is applied consistently, not that MRC captures human-perceived coherence: the rater was shown each step’s framework label and given the scoring rule that MRC’s stability and drift components implement. Appendix E reports the human comparison, where raters saw no framework labels.

Category	MRC (mean $\pm$ std)	$n$
Overall	$0.388\pm 0.175$	3,596
Single-framework	$0.698\pm 0.024$	621
Bounce	$0.380\pm 0.074$	2,168
High-entropy	$0.172\pm 0.030$	807

Appendix

Table 25: MRC summary across trajectory categories. Categories are defined by dominant-framework transition patterns (see Section B.9 ): single-framework = one framework throughout; bounce = framework switches then returns; high-entropy = frequent transitions among multiple frameworks.

Figure 19: MRC against Scoring LLM coherence ratings, over the 113 rated items whose text is a single model’s trajectory (

r=0.961

p<0.0001

). Colors indicate trajectory category; the dashed line is the least-squares fit. The rater was shown each step’s framework label, so this measures consistency of the taxonomy rather than agreement with perceived coherence.

Figure 20: Distribution of MRC scores across trajectory categories. Violin and box for each category over all 3,596 trajectories. Single-framework trajectories concentrate at high MRC (

\mu=0.698

), bounce at intermediate (

\mu=0.380

), high-entropy at low (

\mu=0.172

), with little overlap between the outer two.

Figure 21: MRC score versus Framework Drift Rate (FDR). MRC declines monotonically with drift (Pearson

r=-0.969

). Violins show the MRC distribution at each attainable FDR value over all 3,596 trajectories; the line joins the per-level medians (0.691, 0.485, 0.368, 0.178).

Figure 22: Persuasion resistance by trajectory framework consistency. Flip rates across steering strengths for framework-consistent versus drifting trajectories, split on FDR rather than on MRC. Unstable trajectories show consistently higher flip rates across all steering strengths, with 1.29 $\times$ susceptibility ratio at baseline.

Figure 23: MRC component contributions by trajectory category. Mean of each component by category over all 3,596 trajectories. Stability and drift separate the categories cleanly (Stability 1.000 / 0.647 / 0.471; Drift 1.000 / 0.434 / 0.000), whereas the variance component is uniformly small (0.095 / 0.060 / 0.045) and contributes little discrimination.

Model	Dataset	Stable	Unstable	Diff
Qwen	ETHICS	63.9%	51.1%	+12.8
GPT-5	Moral St.	99.0%	98.7%	+0.2
GPT-5	Social Ch.	47.6%	47.7%	$-$ 0.0
Llama	ETHICS	60.6%	62.6%	$-$ 2.0
Qwen	Social Ch.	50.0%	52.1%	$-$ 2.1
Llama	Social Ch.	54.1%	57.0%	$-$ 2.9

Appendix

Table 26: Stability-accuracy by model

\times

dataset stratum. Only 2/9 strata show positive difference (stable

>

unstable).

Model	Stable	Unstable	Diff	Direction
GPT-5	78.4%	61.0%	+17.5	Expected
Qwen	32.1%	30.0%	+2.1	Expected
Llama	45.0%	55.5%	$-$ 10.5	Opposite
Overall	54.8%	49.6%	+5.1	–

Appendix

Table 27: Stability-accuracy on disagreement cases (

n=563

). GPT-5 shows strong expected effect; Llama shows opposite pattern.

Framework	Score
Kantian Deontology
Benthamite Act Utilitarianism
Aristotelian Virtue Ethics
Scanlonian Contractualism
Gauthierian Contractarianism
Total	100

Appendix

Table 51

Metric	Score
Justified	___ (true/false)
Confidence	___ (0–100)

Appendix

Table 52

Explore similar work

Frame-Conditioned Moral Computation in LLaMA 3.1-8B-Instruct: A Mechanistic Interpretability Audit of Ethical Reasoning
Jun 13, 2026Ali Dasdan, Manan Shah, W. Russell Neuman +3Moral Reasoning Task Framing
1KD Consulting, CA, USA · 2New York University, NY, USA
Normative Robustness as a Frontier for Non-Verifiable Reasoning in LLMs
Jun 10, 2026Elizaveta Tennant, Benjamin Henke, Anita Keshmirian +5Moral Reasoning Large Language Model Reliability
Google DeepMind · Imperial College London · Institute of Philosophy, School of Advanced Study, University of London +2
How Does Thinking Mode Change LLM Moral Judgments? A Controlled Instant-vs-Thinking Comparison Across Five Frontier Models
May 6, 2026Sai Sourabh MadurMoral Reasoning Reasoning Skills
Independent researcher (ex-Meta)

Jun 13, 2026, cs.AIJ/K move · Enter open · S save

Frame-Conditioned Moral Computation in LLaMA 3.1-8B-Instruct: A Mechanistic Interpretability Audit of Ethical Reasoning

Ali Dasdan, Manan Shah, W. Russell Neuman, Chad Coleman+2

1KD Consulting, CA, USA · 2New York University, NY, USA

Moral Reasoning Task Framing Mechanistic Interpretability Model Auditing Reasoning Skills

Open paper AskPDF arXiv

Behavioral audits of Large Language Models on moral prompts measure what the model says, not the internal computation producing it. We use Transluce, an AI-driven mechanistic-interpretability platform, to examine LLaMA 3.1-8B-Instruct on 54 moral prompts in four batteries: 17 dilemmas, policy, and meta-ethical questions (B1); 6 role-playing scenarios (B3); and a controlled trolley contrast varying the switching mechanism with people fixed (B4, 15 prompts) or identity attributes with mechanism fixed (B5, 16 prompts). Two complementary metric families, five cluster-level metrics and a six-metric neuron-level panel, converge on a Situational Anchor Effect: domain-specific representations dominate the top of the activation list across every battery. The model's ethics-labeled capacity stays essentially constant; its salience (rank, priority, top-of-list presence) is highly sensitive to the interpretive frame the prompt selects. The B4-vs-B5 contrast confirms the model attends to whichever surface feature varies: aggregate ethics metrics are indistinguishable, but the dominant non-ethics distractor mirrors the design. A multi-temperature audit identifies a candidate ethics neuron (L16/N3837) stable across temperatures; a cross-model behavioral proxy on two frontier models yields preliminary evidence of divergence in self-reported moral focus, consistent with an Alignment Wrapper in which RLHF re-orders surface text without removing underlying domain-first frames. We unify these as Frame-Conditioned Moral Computation: the prompt's surface vocabulary selects a feature manifold, and the moral conclusion is downstream of that selection. Behavioral alignment must be supplemented by Mechanistic Alignment: a research program asking whether ethics-related features can be shown causally privileged under controlled frame variation, not merely loud in the explanation.

Abstract

Explore similar work

Frame-Conditioned Moral Computation in LLaMA 3.1-8B-Instruct: A Mechanistic Interpretability Audit of Ethical Reasoning

Normative Robustness as a Frontier for Non-Verifiable Reasoning in LLMs

How Does Thinking Mode Change LLM Moral Judgments? A Controlled Instant-vs-Thinking Comparison Across Five Frontier Models