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

2026-06-13

AI Summary

This study addresses a critical gap in the moral auditing of large language models (LLMs), which typically focuses on output content while neglecting internal computational mechanisms. Leveraging the Transluce platform, the authors conduct a mechanistic interpretability analysis of LLaMA-3.1-8B-Instruct across 54 moral prompts, introducing the concept of “framing-conditioned moral computation.” They demonstrate that the model’s ethical judgments are governed by prompt-driven feature manifolds rather than shifts in intrinsic ethical capacity. Through clustering, neuron-level multi-metric analysis, multi-temperature testing, and cross-model behavioral proxy methods, they identify a contextual anchoring effect—where domain-specific representations consistently dominate activation—and pinpoint a stable candidate ethical neuron, L16/N3837. Furthermore, attention shifts with prompt variables while ethical metrics remain constant, though interference sources vary dynamically. The work advocates shifting from behavioral alignment to mechanistic alignment in AI ethics.

Abstract

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.

Problem

• mechanistic interpretability
• moral reasoning
• frame effect
• large language models
• behavioral alignment

Innovation

• Frame-Conditioned Moral Computation
• Mechanistic Interpretability
• Situational Anchor Effect
• Mechanistic Alignment
• Ethics Neuron