mlsanityData to 14 Oct 2026

Explaining Intrinsic Moral Self-Correction with Mechanistic Interpretability

arXiv 2505.11924 ↗First posted 17 May 2025cs.CL cs.AI cs.LGcited by 2

On the map: Mechanistic interpretability → Probing deception, bias and beliefs

Abstract (arXiv)

Intrinsic moral self-correction refers to the phenomenon where a language model refines its ethical judgments or aligns its outputs purely through prompting. While effective across diverse tasks, its mechanism remains unclear. We hypothesize intrinsic moral self-correction functions by steering hidden representations along interpretable latent directions. Evaluating six LLMs across four morality-related tasks, we demonstrate that the representation shifts induced by self-correction prompts align with contrastive steering vectors. This alignment transfers even when the steering vectors are constructed from a disjoint corpus. Notably, when applied via activation addition, these prompt-induced shifts can alter model behavior more effectively than the self-correction prompts and the steering vectors. Our findings suggest representation steering is the mechanistic driver of intrinsic moral self-correction.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Probing deception, bias and beliefs, one of 13 directions in Mechanistic interpretability. Builds on 14 of its 21 references that are in the map. First posted on arXiv 17 May 2025.

Builds on

14 of 21 references.

References · 14

  1. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  2. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  3. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,255
  4. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  5. 2023Steering Language Models With Activation Engineeringcited by 922
  6. 2020Aligning AI With Shared Human Valuescited by 723
  7. 2023Zephyr: Direct Distillation of LM Alignmentcited by 506
  8. 2024When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction of LLMscited by 240
  9. 2020Moral Stories: Situated Reasoning about Norms, Intents, Actions, and their Consequencescited by 135
  10. 2024Confidence Matters: Revisiting Intrinsic Self-Correction Capabilities of Large Language Modelscited by 47
  11. 2024A Theoretical Understanding of Self-Correction through In-context Alignmentcited by 44
  12. 2025Improving Reasoning Performance in Large Language Models via Representation Engineeringcited by 23
2 more
  1. 2024On the Intrinsic Self-Correction Capability of LLMs: Uncertainty and Latent Conceptcited by 16
  2. 2024Intrinsic Self-correction for Enhanced Morality: An Analysis of Internal Mechanisms and the Superficial Hypothesiscited by 15

Built on by

Cited by year

Citing papers per year: 2025: 2.

2 papers.

  1. 2025A Survey of Context Engineering for Large Language Modelscited by 126
  2. 2025Unveiling the Latent Directions of Reflection in Large Language Modelscited by 2

Similar, not cited

Close in wording, no citation either way

  1. 2024Self-correction is Not An Innate Capability in Language Models
  2. 2023The Capacity for Moral Self-Correction in Large Language Models
  3. 2024Smaller Large Language Models Can Do Moral Self-Correction
  4. 2026Frame-Conditioned Moral Computation in LLaMA 3.1-8B-Instruct: A Mechanistic Interpretability Audit of Ethical Reasoning
  5. 2026Tracing Moral Foundations in Large Language Models
  6. 2026The Fragility Of Moral Judgment In Large Language Models
  7. 2025Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability
  8. 2026Representational alignment yields generalizable safety in language models