mlsanityData to 14 Oct 2026

SCAR: Sparse Conditioned Autoencoders for Concept Detection and Steering in LLMs

arXiv 2411.07122 ↗First posted 11 Nov 2024cs.CLcited by 7

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

Large Language Models (LLMs) have demonstrated remarkable capabilities in generating human-like text, but their output may not be aligned with the user or even produce harmful content. This paper presents a novel approach to detect and steer concepts such as toxicity before generation. We introduce the Sparse Conditioned Autoencoder (SCAR), a single trained module that extends the otherwise untouched LLM. SCAR ensures full steerability, towards and away from concepts (e.g., toxic content), without compromising the quality of the model's text generation on standard evaluation benchmarks. We demonstrate the effective application of our approach through a variety of concepts, including toxicity, safety, and writing style alignment. As such, this work establishes a robust framework for controlling LLM generations, ensuring their ethical and safe deployment in real-world applications.

Key facts. Cited by 7 papers in the map (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. Builds on 25 of its 34 references that are in the map. First posted on arXiv 11 Nov 2024.

Builds on

25 of 34 references.

Closest ancestors · 1

Same direction, newest first

  1. 2024ALERT: A Comprehensive Benchmark for Assessing Large Language Models' Safety through Red Teaming6 Apr

Other references · 24

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  4. 2021On the Opportunities and Risks of Foundation Modelscited by 4,033
  5. 2016Pointer Sentinel Mixture Modelscited by 3,096
  6. 2023A Survey of Large Language Modelscited by 2,576
  7. 2022Emergent Abilities of Large Language Modelscited by 2,512
  8. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  9. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  10. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,255
  11. 2023A Survey on Evaluation of Large Language Modelscited by 1,242
  12. 2023Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversationscited by 1,207
12 more
  1. 2021Ethical and social risks of harm from Language Modelscited by 966
  2. 2019Plug and Play Language Models: A Simple Approach to Controlled Text Generationcited by 799
  3. 2024Scaling and evaluating sparse autoencoderscited by 553
  4. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 394
  5. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 300
  6. 2023ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI Conversationcited by 231
  7. 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 179
  8. 2024AEGIS: Online Adaptive AI Content Safety Moderation with Ensemble of LLM Expertscited by 136
  9. 2023Evaluating the Social Impact of Generative AI Systems in Systems and Societycited by 92
  10. 2017Shakespearizing Modern Language Using Copy-Enriched Sequence-to-Sequence Modelscited by 57
  11. 2023PREADD: Prefix-Adaptive Decoding for Controlled Text Generationcited by 29
  12. 2022Revision Transformers: Instructing Language Models to Change their Valuescited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 3.

7 papers.

  1. 2025Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEscited by 14
  2. 2025Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safetycited by 10
  3. 2026Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraintscited by 4
  4. 2025Cross-Layer Discrete Concept Discovery for Interpreting Language Modelscited by 2
  5. 2026Vector Quantized Latent Concepts: A Scalable Alternative to Clustering-Based Concept Discoverycited by 1
  6. 2025Keep Calm and Avoid Harmful Content: Concept Alignment and Latent Manipulation Towards Safer Answerscited by 0
  7. 2026TRACE: Temporal Risk Assessment and Concept Editing for Safer Language Generationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Breaking Bad Tokens: Detoxification of LLMs Using Sparse Autoencoders
  2. 2025Don't Forget It! Conditional Sparse Autoencoder Clamping Works for Unlearning
  3. 2025CRISP: Persistent Concept Unlearning via Sparse Autoencoders
  4. 2025Detoxifying Large Language Models via Autoregressive Reward Guided Representation Editing
  5. 2025Sparse Autoencoder Features for Classifications and Transferability
  6. 2024S-Eval: Towards Automated and Comprehensive Safety Evaluation for Large Language Models
  7. 2025Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework
  8. 2025Measuring and Guiding Monosemanticity