mlsanityData to 7 Oct 2026

Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment

arXiv 2405.13068 ↗First posted 20 May 2024cs.CR cs.AI cs.LGcited by 16

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

also in 1 more
Abstract (arXiv)

Large language models (LLMs) have revolutionized various applications, making robust safety alignment essential to prevent harmful outputs. Current safety alignment techniques, however, harbor inherent vulnerabilities due to their reliance on logit suppression. In this work, we identify critical logit-level vulnerabilities by introducing Semantic-sensitive Alignment and Generation (SSAG), a method designed to systematically manipulate output-layer logits without altering model parameters. Experiments on five popular LLMs show that SSAG exposes harmful responses with a 95% success rate while reducing response time by 86%. VulMine also demonstrates superior attack efficacy, achieving an average ASR of up to 77% against strong defensive mechanisms. These findings reveal crucial weaknesses in existing alignment methods, highlighting an urgent need for improved vulnerability detection and robust safety alignment strategies. Our code is available on github.

Key facts. Cited by 16 papers in the map (data through 7 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 27 of its 28 references that are in the map. First posted on arXiv 20 May 2024.

Builds on

27 of 28 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024Don't Say No: Jailbreaking LLM by Suppressing Refusal25 Apr
  2. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!5 Oct
  3. 2023Universal and Transferable Adversarial Attacks on Aligned Language Models27 Jul

Other references · 24

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  4. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  5. 2023Mistral 7Bcited by 3,065
  6. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,330
  7. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,326
  8. 2024Gemma: Open Models Based on Gemini Research and Technologycited by 889
  9. 2023GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Promptscited by 520
  10. 2024JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Modelscited by 494
  11. 2021Large Dual Encoders Are Generalizable Retrieverscited by 454
  12. 2024Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attackscited by 450
12 more
  1. 2023SmoothLLM: Defending Large Language Models Against Jailbreaking Attackscited by 434
  2. 2023Detecting Language Model Attacks with Perplexitycited by 411
  3. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 377
  4. 2023MasterKey: Automated Jailbreak Across Multiple Large Language Model Chatbotscited by 172
  5. 2024COLD-Attack: Jailbreaking LLMs with Stealthiness and Controllabilitycited by 162
  6. 2024AdvPrompter: Fast Adaptive Adversarial Prompting for LLMscited by 146
  7. 2024Fight Back Against Jailbreaking via Prompt Adversarial Tuningcited by 63
  8. 2024Pandora: Jailbreak GPTs by Retrieval Augmented Generation Poisoningcited by 62
  9. 2024Spectral Editing of Activations for Large Language Model Alignmentcited by 54
  10. 2024Knowledge Graph Enhanced Large Language Model Editingcited by 38
  11. 2024Relative Preference Optimization: Enhancing LLM Alignment through Contrasting Responses across Identical and Diverse Promptscited by 36
  12. 2024MemLLM: Finetuning LLMs to Use An Explicit Read-Write Memorycited by 35

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 9; 2025: 4; 2026: 3.

16 papers.

Sort:

By area:

  1. 2024Bag of Tricks: Benchmarking of Jailbreak Attacks on LLMscited by 47
  2. 2024JAILJUDGE: A Comprehensive Jailbreak Judge Benchmark with Multi-Agent Enhanced Explanation Evaluation Frameworkcited by 44
  3. 2024Emoji Attack: Enhancing Jailbreak Attacks Against Judge LLM Detectioncited by 20
  4. 2024Efficient Detection of Toxic Prompts in Large Language Modelscited by 11
  5. 2025Where to Start Alignment? Diffusion Large Language Model May Demand a Distinct Positioncited by 11
  6. 2024Recent advancements in LLM Red-Teaming: Techniques, Defenses, and Ethical Considerationscited by 8
  7. 2026HoneyTrap: Deceiving Large Language Model Attackers to Honeypot Traps with Resilient Multi-Agent Defensecited by 6
  8. 2024Model-Editing-Based Jailbreak against Safety-aligned Large Language Modelscited by 4
  9. 2024Misinforming LLMs: vulnerabilities, challenges and opportunitiescited by 3
  10. 2024GlitchProber: Advancing Effective Detection and Mitigation of Glitch Tokens in Large Language Modelscited by 3
  11. 2024Gen-AI for User Safety: A Surveycited by 2
  12. 2025Layer-Wise Perturbations via Sparse Autoencoders for Adversarial Text Generationcited by 1
4 more
  1. 2025LLM in the Middle: A Systematic Review of Threats and Mitigations to Real-World LLM-based Systemscited by 1
  2. 2026Talk is (Not) Cheap: A Taxonomy and Benchmark Coverage Audit for LLM Attackscited by 1
  3. 2025Embedding Poisoning: Bypassing Safety Alignment via Embedding Semantic Shiftcited by 0
  4. 2026How Does Prefix Matter in Reasoning Model Tuning?cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Safety Alignment Should Be Made More Than Just a Few Tokens Deep
  2. 2026Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization
  3. 2026LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety
  4. 2025Decoding One Safety Trigger Token for Balancing Safety and Usability in Large Language Models
  5. 2026Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation Trajectories
  6. 2026LLM Safety Alignment in Low-Resource Languages: A Systematic Literature Review
  7. 2025LLM-Safety Evaluations Lack Robustness
  8. 2024Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching