# Robust Harmful Features Under Jailbreak Attacks: Mechanistic Evidence from Attention Head Specialization in Large Language Models

arXiv: https://arxiv.org/abs/2606.28153 · HTML: https://mlsanity.com/paper/2606.28153/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 34 of its 46 references that are in the map.
- First posted on arXiv 26 Jun 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)
- LLM safety and security / [Jailbreaking](https://mlsanity.com/safety/jailbreaking/)

## Builds on (most cited first, 30 of 34)

- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Jailbreaking Black Box Large Language Models in Twenty Queries](https://mlsanity.com/paper/2310.08419/) (2023)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [Progress measures for grokking via mechanistic interpretability](https://mlsanity.com/paper/2301.05217/) (2023)
- [Towards Automated Circuit Discovery for Mechanistic Interpretability](https://mlsanity.com/paper/2304.14997/) (2023)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- [SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks](https://mlsanity.com/paper/2310.03684/) (2023)
- [Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation](https://mlsanity.com/paper/2310.06987/) (2023)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [Detecting Language Model Attacks with Perplexity](https://mlsanity.com/paper/2308.14132/) (2023)
- [Mechanistic Interpretability for AI Safety -- A Review](https://mlsanity.com/paper/2404.14082/) (2024)
- [A Holistic Approach to Undesired Content Detection in the Real World](https://mlsanity.com/paper/2208.03274/) (2022)
- [Improving Alignment and Robustness with Circuit Breakers](https://mlsanity.com/paper/2406.04313/) (2024)
- [SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decoding](https://mlsanity.com/paper/2402.08983/) (2024)
- [OR-Bench: An Over-Refusal Benchmark for Large Language Models](https://mlsanity.com/paper/2405.20947/) (2024)
- [ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI Conversation](https://mlsanity.com/paper/2310.17389/) (2023)
- [SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models](https://mlsanity.com/paper/2402.05044/) (2024)
- [WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models](https://mlsanity.com/paper/2406.18510/) (2024)
- [Programming Refusal with Conditional Activation Steering](https://mlsanity.com/paper/2409.05907/) (2024)
- [Safety Layers in Aligned Large Language Models: The Key to LLM Security](https://mlsanity.com/paper/2408.17003/) (2024)
- [Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails](https://mlsanity.com/paper/2501.09004/) (2025)
- [Llama Guard 3 Vision: Safeguarding Human-AI Image Understanding Conversations](https://mlsanity.com/paper/2411.10414/) (2024)
- [The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence](https://mlsanity.com/paper/2502.17420/) (2025)
- [How Alignment and Jailbreak Work: Explain LLM Safety through Intermediate Hidden States](https://mlsanity.com/paper/2406.05644/) (2024)
- [Copy Suppression: Comprehensively Understanding an Attention Head](https://mlsanity.com/paper/2310.04625/) (2023)

## Built on by (most cited first, 1 of 1)

- [Measuring the Wrong Thing: Internal Harmfulness Scores Anti-Rank Successful Jailbreaks](https://mlsanity.com/paper/2608.09624/) (2026)

## Abstract (arXiv)

Jailbreak attacks bypass LLM safety alignment, yet their mechanisms remain poorly understood. We provide evidence that attacks do not comprehensively eliminate safety features, but instead selectively suppress specific attention heads. We identify two functionally differentiated types: Adversarially Compromised Heads (ACHs) concentrated in early layers, which are suppressed under attacks, and Safety-Aligned Heads (SAHs) in mid-layers, which maintain robust activations even when attacks succeed. Ablation studies support the causal role of ACHs and the contribution of SAHs to robust activations: suppressing a small number of ACHs is sufficient to induce jailbreak-like behavior on normally refused inputs, while removing SAHs substantially weakens mid-layer safety activations. Token-level attribution further shows that ACH suppression is driven specifically by attack-template tokens, providing a mechanistic account of why attacks can bypass refusal decisions through ACH suppression while leaving internal safety signals sustained by SAHs -- a phenomenon we term Robust Harmful Features. To validate the practical significance of this robustness, we show that simply reading these persistent activations -- without any training -- yields competitive aggregate detection performance with strong adversarial robustness.