Mechanistic Interpretability

Overview

The field seeks to explain how a model’s internal computations produce particular outputs, rather than relying only on input-output behavior. Researchers use sparse autoencoders and transcoders to identify features, and probes, attribution graphs, activation patching, ablation, and steering to trace or test their roles. Studies examine language-model tasks such as factual recall, reasoning, and refusal, as well as vision, audio, and scientific models. A central challenge is that a signal can be easy to decode without being causally used, while circuit explanations and interventions can vary across prompts, models, and analysis choices.

Latest papers

CardsList

427 papers58 papers in the last four weeks, up 23% on the four weeks before. 0.4% of all new papers.

  1. Goal Encoding in LLMs

    Nov 12, 2026Arnab Sen Sharma, Eric Todd, Sheridan Feucht +7Mechanistic Interpretability

  2. Objects as Addresses: Instance Binding in Vision Transformer Attention

    Nov 12, 2026Khalil Sabri, Guillaume-Alexandre Bilodeau, Nicolas Saunier +1Vision TransformerVisual Attention

  3. A Mechanistic Interpretability Study of an Astronomy Foundation Model

    Nov 11, 2026Rachel C. Zhang, Alberto Bietti, Wes Gurnee +2Multimodal Model InterpretabilityMechanistic Interpretability

  4. Mechanistic Interpretability Needs a Logic of Inference

    Nov 10, 2026Zhimin Hu, Chen Jiang, Hanbo Xie +2Mechanistic InterpretabilityNeural Network Interpretability

  5. Interpretability Is Not Model-Specific

    Nov 9, 2026Marcus Armstrong, Navid Ayoobi, Pradham Mummaleti +2Interpretable MLMechanistic Interpretability

  6. Represented but Not Used: Value-Policy Dissociation in LLM Agents

    Nov 7, 2026Yuzhang Luo, Mohamed Aghzal, Liangming Pan +1LLM AgentsMechanistic Interpretability

  7. Detecting and Explaining Robot Policy Failures from Sparse Internal Features

    Nov 7, 2026Riad Ahmed, Sujosh Nag, Moniruzzaman Akash +1Robot Failure DetectionMechanistic Interpretability

  8. Same Behavior, Different Mechanism: What Finite Experiments Can Certify About Integrated Information

    Nov 5, 2026Linqian Wu, Diaoxuan HeCausal IdentifiabilityMechanistic Interpretability

  9. Understanding AI from the Inside: The State of Mechanistic Interpretability for AI Safety

    Nov 5, 2026Fazl Barez, Tung-Yu Wu, Hunar Batra +24Mechanistic InterpretabilityAI Safety

  10. The Intrusion Pathway: How Non-Evidential Context Degrades Language Model Performance

    Nov 5, 2026Dikshant Kukreja, Gautam Gupta, Avni Mittal +8Language Model RobustnessMechanistic Interpretability

  11. When Every Choice Is Correct: A Causal Study of Numerical Preference in Language Models

    Nov 4, 2026Hongnian Huang, Hongliang LiuMechanistic InterpretabilityCircuits in Language Models

  12. From Activation to Intervention: Interpreting Sparse Visual Features through Reconstruction

    Nov 3, 2026Jianyang Gu, Jacob Beattie, Mridul Khurana +7Neural Network InterpretabilityMechanistic Interpretability

  13. Whose Effect Is It? How Objectives Inflate Learned Subspace Interventions

    Nov 3, 2026Toygar Tanyel, Serkan AyvazMechanistic InterpretabilityExplainability Evaluation

  14. Where Do Distilled Students Store Thinking?

    Nov 3, 2026Taeyeong Kim, Jaemin Jung, Ahhyun Kim +1CoT DistillationKnowledge Distillation

  15. A balloon effect: ban one channel and a hidden trait finds another

    Nov 2, 2026Keuntae Kim, Yong Suk ChoiSubliminal LearningCausal Interventions in Language Models

  16. Multi-Agent Collusion and the J-Space

    Nov 1, 2026Marco Molinari, Gabriele BettineschiMulti-Agent System SecurityMulti-Agent LLMs

  17. Opening the Black Box of Neural Auction Mechanisms

    Nov 1, 2026Li Kong, Zilong Liu, Haoyang Zhou +2Mechanism DesignMechanistic Interpretability

  18. Who Asks, Who Answers: Linear Attention Builds the Query in Hybrid Language Models

    Nov 1, 2026Wenqi Zeng, Chen Feng, YuAng ChenKnowledge Retrieval in Language ModelsHybrid Language Models

  19. Chain-of-Thought Oversight Requires Separate Evidence for Faithfulness and Monitorability

    Nov 1, 2026Sichao Li, Sai Ma, Xiyang Hu +2CoT MonitoringFaithfulness of Language Model Reasoning

  20. Bridging Singular Learning Theory and Mechanistic Interpretability at the Billion-Parameter Scale

    Oct 31, 2026Ethan Ji, Wei Wang, Yizhou SunMechanistic InterpretabilitySingular Learning Theory

  21. Editing the Goal: Per-Token SAE Steering and Its Depth Limits in π0.5π_{0.5}

    Oct 31, 2026Rocio Perales Valdes, Animesh GargActivation SteeringMechanistic Interpretability

  22. Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries

    Oct 30, 2026Shariar Kabir, Hong Kiat Tan, Swastik Agrawal +2LLM InterpretabilityMechanistic Interpretability

  23. Nonconformal Distortion and Local Orientation Reversal in the Fourier Circuit of a Modular-Addition Transformer

    Oct 30, 2026Eduardo Fuentes, Gonzalo Riera, Héctor Allende-CidNeural Representation GeometryMechanistic Interpretability

  24. Mechanistic Interpretability for End-to-End Self-Driving

    Oct 30, 2026Manav Gagvani, Benjamin Namikas, Sivamurugan Velmurugan +6Mechanistic InterpretabilityAutonomous Driving