# Applying sparse autoencoders to unlearn knowledge in language models

arXiv: https://arxiv.org/abs/2410.19278 · HTML: https://mlsanity.com/paper/2410.19278/

## Key facts

- Cited by 59 papers in the map (data through 14 Oct 2026).
- Placed in Machine unlearning, one of 12 directions in LLM safety and security.
- Builds on 10 of its 17 references that are in the map.
- First posted on arXiv 25 Oct 2024.

## Placed in

- LLM safety and security / [Machine unlearning](https://mlsanity.com/safety/machine-unlearning/)
- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 10 of 10)

- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Survey of Hallucination in Natural Language Generation](https://mlsanity.com/paper/2202.03629/) (2022)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2](https://mlsanity.com/paper/2408.05147/) (2024)
- [Who's Harry Potter? Approximate Unlearning in LLMs](https://mlsanity.com/paper/2310.02238/) (2023)
- [Improving Alignment and Robustness with Circuit Breakers](https://mlsanity.com/paper/2406.04313/) (2024)
- [An Adversarial Perspective on Machine Unlearning for AI Safety](https://mlsanity.com/paper/2409.18025/) (2024)

## Built on by (most cited first, 30 of 59)

- [SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability](https://mlsanity.com/paper/2503.09532/) (2025)
- [Are Sparse Autoencoders Useful? A Case Study in Sparse Probing](https://mlsanity.com/paper/2502.16681/) (2025)
- [Steering Language Model Refusal with Sparse Autoencoders](https://mlsanity.com/paper/2411.11296/) (2024)
- [SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoders](https://mlsanity.com/paper/2501.18052/) (2025)
- [A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models](https://mlsanity.com/paper/2503.05613/) (2025)
- [Enhancing Automated Interpretability with Output-Centric Feature Descriptions](https://mlsanity.com/paper/2501.08319/) (2025)
- [Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning](https://mlsanity.com/paper/2507.16795/) (2025)
- [Understanding the Repeat Curse in Large Language Models from a Feature Perspective](https://mlsanity.com/paper/2504.14218/) (2025)
- [Steering off Course: Reliability Challenges in Steering Language Models](https://mlsanity.com/paper/2504.04635/) (2025)
- [SAEs Can Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMs](https://mlsanity.com/paper/2504.08192/) (2025)
- [Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models](https://mlsanity.com/paper/2605.11887/) (2026)
- [Position: Use Sparse Autoencoders to Discover Unknowns](https://mlsanity.com/paper/2506.23845/) (2025)
- [Model Unlearning via Sparse Autoencoder Subspace Guided Projections](https://mlsanity.com/paper/2505.24428/) (2025)
- [Tracking the Feature Dynamics in LLM Training: A Mechanistic Study](https://mlsanity.com/paper/2412.17626/) (2024)
- [SAUCE: Selective Concept Unlearning in Vision-Language Models with Sparse Autoencoders](https://mlsanity.com/paper/2503.14530/) (2025)
- [Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models](https://mlsanity.com/paper/2505.15634/) (2025)
- [Precise In-Parameter Concept Erasure in Large Language Models](https://mlsanity.com/paper/2505.22586/) (2025)
- [SoK: Machine Unlearning for Large Language Models](https://mlsanity.com/paper/2506.09227/) (2025)
- [CRISP: Persistent Concept Unlearning via Sparse Autoencoders](https://mlsanity.com/paper/2508.13650/) (2025)
- [Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations](https://mlsanity.com/paper/2502.18147/) (2025)
- [AbsTopK: Rethinking Sparse Autoencoders For Bidirectional Features](https://mlsanity.com/paper/2510.00404/) (2025)
- [Hypothesis-Driven Feature Manifold Analysis in LLMs via Supervised Multi-Dimensional Scaling](https://mlsanity.com/paper/2510.01025/) (2025)
- [AudioSAE: Towards Understanding of Audio-Processing Models with Sparse AutoEncoders](https://mlsanity.com/paper/2602.05027/) (2026)
- [EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Models](https://mlsanity.com/paper/2504.15133/) (2025)
- [DLM-Scope: Mechanistic Interpretability of Diffusion Language Models via Sparse Autoencoders](https://mlsanity.com/paper/2602.05859/) (2026)
- [Understanding Emergent Misalignment via Feature Superposition Geometry](https://mlsanity.com/paper/2605.00842/) (2026)
- [Don't Forget It! Conditional Sparse Autoencoder Clamping Works for Unlearning](https://mlsanity.com/paper/2503.11127/) (2025)
- [Towards Interpretability Without Sacrifice: Faithful Dense Layer Decomposition with Mixture of Decoders](https://mlsanity.com/paper/2505.21364/) (2025)
- [Signal in the Noise: Polysemantic Interference Transfers and Predicts Cross-Model Influence](https://mlsanity.com/paper/2505.11611/) (2025)
- [Generative Model Unlearning: A Survey through Target Events, Unlearning Operators, and Evaluation Protocols](https://mlsanity.com/paper/2507.19894/) (2025)

## Abstract (arXiv)

We investigate whether sparse autoencoders (SAEs) can be used to remove knowledge from language models. We use the biology subset of the Weapons of Mass Destruction Proxy dataset and test on the gemma-2b-it and gemma-2-2b-it language models. We demonstrate that individual interpretable biology-related SAE features can be used to unlearn a subset of WMDP-Bio questions with minimal side-effects in domains other than biology. Our results suggest that negative scaling of feature activations is necessary and that zero ablating features is ineffective. We find that intervening using multiple SAE features simultaneously can unlearn multiple different topics, but with similar or larger unwanted side-effects than the existing Representation Misdirection for Unlearning technique. Current SAE quality or intervention techniques would need to improve to make SAE-based unlearning comparable to the existing fine-tuning based techniques.