mlsanityData to 14 Oct 2026

SABER: Uncovering Vulnerabilities in Safety Alignment via Cross-Layer Residual Connection

arXiv 2509.16060 ↗First posted 19 Sep 2025cs.LG cs.CLcited by 1

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

also in 1 more
Abstract (arXiv)

Large Language Models (LLMs) with safe-alignment training are powerful instruments with robust language comprehension capabilities. These models typically undergo meticulous alignment procedures involving human feedback to ensure the acceptance of safe inputs while rejecting harmful or unsafe ones. However, despite their massive scale and alignment efforts, LLMs remain vulnerable to jailbreak attacks, where malicious users manipulate the model to produce harmful outputs that it was explicitly trained to avoid. In this study, we find that the safety mechanisms in LLMs are predominantly embedded in the middle-to-late layers. Building on this insight, we introduce a novel white-box jailbreak method, SABER (Safety Alignment Bypass via Extra Residuals), which connects two intermediate layers s and e such that s < e, through a residual connection. Our approach achieves a 51% improvement over the best-performing baseline on the HarmBench test set. Furthermore, SABER induces only a marginal shift in perplexity when evaluated on the HarmBench validation set. The source code is publicly available at https://github.com/PalGitts/SABER.

Key facts. Cited by 1 paper 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 32 of its 52 references that are in the map. First posted on arXiv 19 Sep 2025.

Builds on

32 of 52 references.

Closest ancestors · 4

Same direction, newest first

  1. 2023Jailbreak and Guard Aligned Language Models with Only Few In-Context Demonstrations10 Oct
  2. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!5 Oct
  3. 2023Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models4 Oct
  4. 2023Universal and Transferable Adversarial Attacks on Aligned Language Models27 Jul

Other references · 28

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  3. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,886
  4. 2019HellaSwag: Can a Machine Really Finish Your Sentence?cited by 4,100
  5. 2017Deep reinforcement learning from human preferencescited by 3,912
  6. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  7. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,420
  8. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  9. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  10. 2022BioGPT: Generative Pre-trained Transformer for Biomedical Text Generation and Miningcited by 538
  11. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 435
  12. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 387
16 more
  1. 2023DeepInception: Hypnotize Large Language Model to Be Jailbreakercited by 367
  2. 2023Emergent Linear Representations in World Models of Self-Supervised Sequence Modelscited by 319
  3. 2023A Wolf in Sheep's Clothing: Generalized Nested Jailbreak Prompts can Fool Large Language Models Easilycited by 259
  4. 2023Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attackscited by 194
  5. 2023Tree-Structured Parzen Estimator: Understanding Its Algorithm Components and Their Roles for Better Empirical Performancecited by 147
  6. 2024A Comprehensive Study of Jailbreak Attack versus Defense for Large Language Modelscited by 106
  7. 2024JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Modelscited by 70
  8. 2023BadCLIP: Trigger-Aware Prompt Learning for Backdoor Attacks on CLIPcited by 67
  9. 2023CleanCLIP: Mitigating Data Poisoning Attacks in Multimodal Contrastive Learningcited by 63
  10. 2022Detecting and Understanding Harmful Memes: A Surveycited by 58
  11. 2023The Janus Interface: How Fine-Tuning in Large Language Models Amplifies the Privacy Riskscited by 36
  12. 2023Make Them Spill the Beans! Coercive Knowledge Extraction from (Production) LLMscited by 27
  13. 2022DISARM: Detecting the Victims Targeted by Harmful Memescited by 20
  14. 2023Characterizing the Entities in Harmful Memes: Who is the Hero, the Villain, the Victim?cited by 14
  15. 2024SAFE-MEME: Structured Reasoning Framework for Robust Hate Speech Detection in Memescited by 4
  16. 2024Meme Trojan: Backdoor Attacks Against Hateful Meme Detection via Cross-Modal Triggerscited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Omni-Safety under Cross-Modality Conflict: Vulnerabilities, Dynamics Mechanisms and Efficient Alignmentcited by 6

Similar, not cited

Close in wording, no citation either way

  1. 2024Safety Alignment Should Be Made More Than Just a Few Tokens Deep
  2. 2025Decoding One Safety Trigger Token for Balancing Safety and Usability in Large Language Models
  3. 2025Playing the Fool: Jailbreaking LLMs and Multimodal LLMs with Out-of-Distribution Strategy
  4. 2025TRIDENT: Enhancing Large Language Model Safety with Tri-Dimensional Diversified Red-Teaming Data Synthesis
  5. 2025Safety Alignment Should Be Made More Than Just A Few Attention Heads
  6. 2024Emulated Disalignment: Safety Alignment for Large Language Models May Backfire!
  7. 2026Safety Is Not Universal: The Selective Safety Trap in LLM Alignment
  8. 2025Targeting Alignment: Extracting Safety Classifiers of Aligned LLMs