paper
referenced-only
2023
paper:a-jailbroken-how-does-llm-safety-training-2023Jailbroken: How does LLM safety training fail?
ByA. Wei·N. Haghtalab·J. Steinhardt
Related work— refs + corpus + external arXiv
Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.
- Different Paths to Harmful Compliance: Behavioral Side Effects and Mechanistic Divergence Across LLM JailbreaksMd Rysul Kabir and Zoran Tiganj2026≈ 75%
- The Struggle Between Continuation and Refusal: A Mechanistic Analysis of the Continuation-Triggered Jailbreak in LLMsZhen Yang, Ping Jian, Xinyue Zhang, Zhongbin Guo, Chengzhi Li Yonghong Deng2026≈ 73%
- SafeDream: Safety World Model for Proactive Early Jailbreak DetectionWeikai Lin, Yada Zhu, Song Wang Bo Yan2026≈ 72%
- Safety is Not Only About Refusal: Reasoning-Enhanced Fine-tuning for Interpretable LLM SafetyMiao Li, William Han, Yihang Yao, Zhepeng Cen, Ding Zhao Yuyou Zhang2025≈ 71%
- Revisiting Jailbreaking for Large Language Models: A Representation Engineering PerspectiveZhenghua Wang, Wenhao Liu, Muling Wu, Shihan Dou, Changze Lv, Xiaohua Wang, Xiaoqing Zheng, Xuanjing Huang Tianlong Li2025≈ 71%
- Breaking Bad: Interpretability-Based Safety Audits of State-of-the-Art LLMsRamneet Kaur, Colin Samplawski, Manoj Acharya, Anirban Roy, Daniel Elenius, Brian Matejek, Adam D. Cobb, Susmit Jha Krishiv Agarwal2026≈ 71%
- A Representation Engineering Perspective on the Effectiveness of Multi-Turn JailbreaksMark Russinovich, Ahmed Salem, Santiago Zanella-Beguelin, Daniel Jones, Giorgio Severi, Eugenia Kim, Keegan Hines, Amanda Minnich, Yonatan Zunger, Ram Shankar Siva Kumar Blake Bullwinkel2025≈ 70%
- ≈ 70%
- Automated Framework to Evaluate and Harden LLM System Instructions against Encoding AttacksDiptisha Samanta, Reza Soosahabi Anubhab Sahu2026≈ 69%
- Why Not Act on What You Know? Unleashing Safety Potential of LLMs via Self-Aware Guard EnhancementJun Kuang, Zongyu Wang, Xuezhi Cao, Xunliang Cai, Jiajun Chen, Shujian Huang Peng Ding2025≈ 69%
- Tracing the Dynamics of Refusal: Exploiting Latent Refusal Trajectories for Robust Jailbreak DetectionChe Wang, Wei Yang Bryan Lim, Jianbo Gao, Zhong Chen Xulin Hu2026≈ 69%
- TRYLOCK: Defense-in-Depth Against LLM Jailbreaks via Layered Preference and Representation EngineeringScott Thornton2026≈ 69%
- A Mixture of Linear Corrections Generates Secure CodeRavi Mangal, Terry Zhuo, Matt Fredrikson, Corina S. Pasareanu Weichen Yu2025≈ 69%
- Sell Me This Stock: Unsafe Recommendation Drift in LLM AgentsAdriano Koshiyama, Sahan Bulathwela, Maria Perez-Ortiz Zekun Wu2026≈ 69%
- Probing the Robustness of Large Language Models Safety to Latent PerturbationsKexin Huang, Zongqi Wang, Yixu Wang, Jie Li, Yuanqi Yao, Yang Yao, Yujiu Yang, Yan Teng, Yingchun Wang Tianle Gu2025≈ 68%
- SelfGrader: Stable Jailbreak Detection for Large Language Models using Token-Level LogitsRui Hu, Olivera Kotevska, Jiahao Xu Zikai Zhang2026≈ 68%
- ≈ 68%
- Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluationsin corpus2026≈ 67%
- Probe-Based Data Attribution: Surfacing and Mitigating Undesirable Behaviors in LLM Post-Trainingin corpus2026≈ 67%
- ≈ 66%
- Testing the Limits of Truth Directions in LLMsin corpus2026≈ 66%
- ≈ 65%
- ≈ 64%
- Why Larger Models Learn More: Effects of Capacity, Interference, and Rare-Task Retentionin corpus2026≈ 64%
- Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signalin corpus2026≈ 64%
- Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agentsin corpus2026≈ 64%
- ≈ 64%
- ≈ 64%
- ≈ 64%
- Alignment faking in large language modelsin corpus2024≈ 64%
Similar preprints — Semantic Scholar
Cited by (3)
- Understanding and Steering Llama 3 with Sparse Autoencoders - Goodfire
A sparse autoencoder trained on layer-19 residual-stream activations of Llama-3-8B-Instruct, using LMSYS-Chat-1M, yields features that can be labeled automatically, ranked by causal relevance, and use
- Unveiling the Latent Directions of Reflection in Large Language Models
Reflection in LLMs corresponds to a recoverable latent direction in activation space, not merely a behavioral artifact of prompt engineering. Working with Qwen2.5-3B and Gemma3-4B-IT on the adversaria
- Probe-Based Data Attribution: Surfacing and Mitigating Undesirable Behaviors in LLM Post-Training
Probe-based data attribution, introduced here as a method for surfacing and mitigating undesirable post-training behaviors, reduces harmful compliance in OLMo 2 7B by 63% through datapoint filtering a