# Causally Grounded Mechanistic Interpretability for LLMs with Faithful Natural-Language Explanations

arXiv: https://arxiv.org/abs/2603.09988 · HTML: https://mlsanity.com/paper/2603.09988/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Knowledge and representations inside LLMs, one of 13 directions in Mechanistic interpretability.
- Builds on 8 of its 13 references that are in the map.
- First posted on arXiv 13 Feb 2026.

## Placed in

- Mechanistic interpretability / [Knowledge and representations inside LLMs](https://mlsanity.com/mechinterp/knowledge-and-representations-inside-llms/)

## Builds on (most cited first, 8 of 8)

- [Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small](https://mlsanity.com/paper/2211.00593/) (2022)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Towards Automated Circuit Discovery for Mechanistic Interpretability](https://mlsanity.com/paper/2304.14997/) (2023)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [ERASER: A Benchmark to Evaluate Rationalized NLP Models](https://mlsanity.com/paper/1911.03429/) (2019)
- [e-SNLI: Natural Language Inference with Natural Language Explanations](https://mlsanity.com/paper/1812.01193/) (2018)
- [Causal Abstractions of Neural Networks](https://mlsanity.com/paper/2106.02997/) (2021)
- [Towards Faithful Natural Language Explanations: A Study Using Activation Patching in Large Language Models](https://mlsanity.com/paper/2410.14155/) (2024)

## Abstract (arXiv)

Mechanistic interpretability identifies internal circuits responsible for model behaviors, yet translating these findings into human-understandable explanations remains an open problem. We present a pipeline that bridges circuit-level analysis and natural language explanations by (i) identifying causally important attention heads via activation patching, (ii) generating explanations using both template-based and LLM-based methods, and (iii) evaluating faithfulness using ERASER-style metrics adapted for circuit-level attribution. We evaluate on the Indirect Object Identification (IOI) task in GPT-2 Small (124M parameters), identifying six attention heads accounting for 61.4% of the logit difference. Our circuit-based explanations achieve 100% sufficiency but only 22% comprehensiveness, revealing distributed backup mechanisms. LLM-generated explanations outperform template baselines by 64% on quality metrics. We find no correlation (r = 0.009) between model confidence and explanation faithfulness, and identify three failure categories explaining when explanations diverge from mechanisms.