paper-with-me

Papers

Sparse Autoencoders are Capable LLM Jailbreak Mitigators

2026-02-12 · Yannick Assogba, Jacopo Cortellazzi, Javier Abad, Pau Rodriguez, Xavier Suau, Arno Blaas arxiv

Jailbreak attacks remain a persistent threat to large language model safety. We propose Context-Conditioned Delta Steering (CC-Delta), an SAE-based defense that identifies jailbreak-relevant sparse features by comparing token-level representations of the same harmful request with and without jailbreak context. Using paired harmful/jailbreak prompts, CC-Delta selects features via statistical testing and applies inference-time mean-shift steering in SAE latent space. Across four aligned instruction-tuned models and twelve jailbreak attacks, CC-Delta achieves comparable or better safety-utility tradeoffs than baseline defenses operating in dense latent space. In particular, our method clearly outperforms dense mean-shift steering on all four models, and particularly against out-of-distribution attacks, showing that steering in sparse SAE feature space offers advantages over steering in dense activation space for jailbreak mitigation. Our results suggest off-the-shelf SAEs trained for interpretability can be repurposed as practical jailbreak defenses without task-specific training.

📄 PDF Abstract BibTeX arXiv:2602.12418

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Graph-Regularized Sparse Autoencoders for LLM Safety Steering

2025-12-07 · Jehyeok Yeon, Federico Cinus, Yifan Wu, Luca Luceri arxiv

Sparse autoencoders (SAEs) are increasingly used to extract activation directions for inference-time steering, but their standard sparsity objective treats latent features as independent. This prior can be poorly matched…

Towards Understanding the Robustness of Sparse Autoencoders

2026-04-20 · Ahson Saiyed, Sabrina Sadiekh, Chirag Agarwal arxiv

Large Language Models (LLMs) remain vulnerable to optimization-based jailbreak attacks that exploit internal gradient structure. While Sparse Autoencoders (SAEs) are widely used for interpretability, their robustness imp…

An Empirical Study of Modular Bias Mitigators and Ensembles

2022-02-01 · Michael Feffer, Martin Hirzel, Samuel C. Hoffman, Kiran Kate 외

There are several bias mitigators that can reduce algorithmic bias in machine learning models but, unfortunately, the effect of mitigators on fairness is often not stable when measured across different data splits. A pop…

Ensemble LearningFairness

ConceptGuard: Neuro-Symbolic Safety Guardrails via Sparse Interpretable Jailbreak Concepts

2025-08-22 · Darpan Aswal, Céline Hudelot arxiv

Large Language Models have found success in a variety of applications. However, their safety remains a concern due to the existence of various jailbreaking methods. Despite significant efforts, alignment and safety fine-…

Causal Front-Door Adjustment for Robust Jailbreak Attacks on LLMs

2026-02-05 · Yao Zhou, Zeen Song, Wenwen Qiang, Fengge Wu 외 arxiv

Safety alignment mechanisms in Large Language Models (LLMs) often operate as latent internal states, obscuring the model's inherent capabilities. Building on this observation, we model the safety mechanism as an unobserv…