Tracing Moral Foundations in Large Language Models

📰 ArXiv cs.AI

Learn how to analyze moral foundations in large language models using Moral Foundations Theory, and why it matters for understanding AI ethics

advanced Published 12 May 2026

Action Steps

Apply Moral Foundations Theory to analyze the moral judgments of large language models
Configure experiments to test the encoding and organization of moral foundations in LLMs
Run comparisons across different model families and scales to identify patterns and trends
Test the expression of moral foundations in LLMs using instruction-tuning and base models
Analyze the results to understand the internal conceptual structure of moral judgments in LLMs

Who Needs to Know This

AI researchers and ethicists can benefit from this knowledge to develop more transparent and accountable language models, while product managers can use it to inform design decisions for AI-powered products

Key Insight

💡 Moral Foundations Theory can be used to analyze and understand the moral judgments of large language models, revealing potential biases and limitations

Key Takeaways

Learn how to analyze moral foundations in large language models using Moral Foundations Theory, and why it matters for understanding AI ethics

Full Article

Title: Tracing Moral Foundations in Large Language Models

Abstract:
arXiv:2601.05437v2 Announce Type: replace-cross Abstract: Large language models often produce human-like moral judgments, but it is unclear whether this reflects an internal conceptual structure or superficial ``moral mimicry.'' Using Moral Foundations Theory (MFT) as an analytic framework, we study how moral foundations are encoded, organized, and expressed across 14 base and instruction-tuned LLMs spanning four model families (Llama, Qwen2.5, Qwen3-MoE, Mistral) and scales from 7B to 70B. We e

Read full paper → ← Back to Reads