Refusal in Language Models Is Mediated by a Single Direction
The DOI matches its Crossref record; 18 of the 18 references it cites were found in a registry or an index.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“ref. 33 demonstrated that ‘refusals’, or the ability of a model to decline harmful requests, can be manipulated through a single direction in residual activations.”
From Training large language models on narrow tasks can lead to broad misalignment (2026) · cites this paper as [33]
In the citing paper
“Importantly, this suggests that harmlessness training (which trains models to refuse to answer harmful requests) may be cheaply undone.”
From Open Problems in Mechanistic Interpretability (2025) · cites this paper as Arditi et al., 2024
In the citing paper
“By understanding the mechanisms of jailbreaking, we can anticipate the means through which a user might bypass existing safeguards (Lee et al., 2025;Arditi et al., 2024).”
From Open Problems in Mechanistic Interpretability (2025) · cites this paper as Arditi et al., 2024
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Cited by
Items 501–600 of 1,000
Activations as Features: Probing LLMs for Generalizable Essay Scoring Representations
Chi, J., Wang, K., Chen, Y., Lin, X., & Xu, Q. (2026) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v40i36.40292 · Cited by 3
HumorReject: Decoupling LLM Safety from Refusal Prefix via a Little Humor
Wu, Z., Gao, H., Luo, J., & Liu, Z. (2026) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v40i44.41140 · Cited by 3
Alignment Reduces Expressed but Not Encoded Gender Bias: A Unified Framework and Study
Bouchouchi, N., Lesot, M., Laugel, T., Renard, X., Marsala, C., & Detyniecki, M. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37654-1_21 · Cited by 3
Maskey, U., Dras, M., & Naseem, U. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1867 · Cited by 4
Simple picture of how output from ChatGPT-like AI shifts from good to bad
Johnson, N. F., & Huo, F. Y. (2026) · PNAS Nexus · doi:10.1093/pnasnexus/pgag148 · Cited by 3
CHASE: Adversarial Red-Blue Teaming for Improving LLM Safety using Reinforcement Learning
Markasserithodi, R., Joshi, A., Li, Y., Singh, I., Yoo, C., & Niu, A. (2026) · International Joint Conference on Natural Language Processing and Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL) · doi:10.48550/arxiv.2606.05523 · Cited by 3
Unlearners Can Lie: Evaluating and Improving Honesty in LLM Unlearning
Gu, R., Du, J., Zhang, Y., & Liu, S. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.548 · Cited by 3
Li, J., Ren, W., & Zhu, T. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.833 · Cited by 3
AdaptiveK: Complexity-Driven Sparse Autoencoders for Interpretable Language Model Representations
Yao, Y., Zhang, H., & Du, M. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.1187 · Cited by 3
From Narrow Unlearning to Emergent Misalignment in LLMs
Mushtaq, E., Ramakrishna, A., Krishna, S., Sahai, S., Goyal, P., Chang, K.-W., Zhang, T., & Gupta, R. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-short.32 · Cited by 3
Probing the Safety Robustness of LLMs in Latent Space
Gu, T., Huang, K., Wang, Z., Wang, Y., Li, J., Wang, X. E., Yao, Y., Yang, Y., Teng, Y., & Wang, Y. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.967 · Cited by 3
Shi, L., Wu, X., Zhao, X., Wang, H., Liu, H., Liu, Y., Xu, L., Wang, L., Xiong, D., & Luo, W. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.283 · Cited by 3
Wendlinger, M., Kowatsch, D., Böttinger, K., & Sperl, P. (2026) · IEEE European Symposium on Security and Privacy · doi:10.1109/eurosp68448.2026.00049 · Cited by 3
Interpreting and Steering for Safe and Correct Code Generation
Yan, H., & Yao, Z. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.923 · Cited by 3
The Trilemma of Truth in Large Language Models
Savcisens, G., & Eliassi‐Rad, T. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.2139/ssrn.5331504 · Cited by 3
Who’s Asking? Simulating Role-Based Questions for Conversational AI Evaluation
Kaur, N., Ayad, H., Jung, H., Mittal, S., Choudhury, M. D., & Mitra, T. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.1769 · Cited by 3
Refusal-Aware Red Teaming: Exposing Inconsistency in Safety Evaluations
Chen, Y., Du, X., Zou, X., Zhao, C. K., Deng, H., Li, H., & Kuang, X. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.49 · Cited by 3
Debias Your Large Multi-Modal Model at Test-Time via Non-Contrastive Visual Attribute Steering
Ratzlaff, N., Olson, M., Hinck, M., Tseng, S.-Y., Lal, V., & Howard, P. (2025) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw69036.2025.00651 · Cited by 3
SAGE: Scalable Ground Truth Evaluations for Large Sparse AutoencodersPreprint
Venhoff, C., Calinescu, A., Torr, P. H. S., & de Witt, C. S. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.07456 · Cited by 3
Toward a Flexible Framework for Linear Representation Hypothesis Using Maximum Likelihood EstimationPreprint
Nguyen, T., & Leng, Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.16385 · Cited by 3
A Generative Approach to LLM Harmfulness Mitigation with Red Flag TokensPreprint
Dobre, D., Mofakhami, M., Xhonneux, S., Schwinn, L., & Gidel, G. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.16366 · Cited by 4
The Structural Safety Generalization ProblemPreprint
Broomfield, J., Gibbs, T., Kosak-Hine, E., Ingebretsen, G., Nasir, T., Zhang, J., Iranmanesh, R., Pieri, S., Rabbany, R., & Pelrine, K. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.09712 · Cited by 3
Shayegani, E., Shahariar, G. M., Abdali, S., Yu, L., Abu‐Ghazaleh, N., & Dong, Y. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2504.03735 · Cited by 3
AlignTree: Efficient Defense Against LLM Jailbreak AttacksPreprint
Goren, G., Katz, S., & Wolf, L. (2025) · AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) · doi:10.48550/arxiv.2511.12217 · Cited by 3
D-STEER: Two-Way Control of DPO-Induced Safety Behavior in Activation SpacePreprint
Raina, S., Aggarwal, S., Dhanda, A., Jain, V., Chadha, A., & Das, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.11838 · Cited by 3
Improving Preference Extraction In LLMs By Identifying Latent Knowledge Through Classifying ProbesPreprint
Maiya, S., Liu, Y., Debnath, R., & Korhonen, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.17755 · Cited by 3
Sycophancy as compositions of Atomic Psychometric TraitsPreprint
Jain, S., Yost, A., & Abdullah, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.19316 · Cited by 3
Knowledge distillation through geometry-aware representational alignmentPreprint
Bhattarai, P., Amjad, M., Zhylko, D., & Alhanai, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.25253 · Cited by 3
Mapping Faithful Reasoning in Language ModelsPreprint
Li, J., Damianou, A., Rosser, J., García, J. L. R., & Palla, K. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.22362 · Cited by 3
Curvature-Aware Safety Restoration In LLMs Fine-Tuning
Bach, T., Nguyen-Tang, T., Nguyen, D., Le, T. M., & Tran, T. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2511.18039 · Cited by 3
Eliciting Chain-of-Thought in Base LLMs via Gradient-Based Representation OptimizationPreprint
Wang, Z., Ma, Y., & Xu, C. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.19131 · Cited by 3
SearchAttack: Red-Teaming LLMs against Knowledge-to-Action Threats under Online Web SearchPreprint
Yan, Y., Sun, S., Li, M., Yang, Z., Zhu, C., Ma, F., Xu, B., Liu, M., & Li, Q. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.04093 · Cited by 3
DriveSafe: A Hierarchical Risk Taxonomy for Safety-Critical LLM-Based Driving AssistantsPreprint
Kumar, A., Tapwal, R., & Maple, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.12138 · Cited by 3
A Behavioral Fingerprint for Large Language Models: Provenance Tracking via Refusal VectorsPreprint
Xu, Z., & Sheng, V. S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.09434 · Cited by 3
AISA: Awakening Intrinsic Safety Awareness in Large Language Models against Jailbreak AttacksPreprint
Song, W., Xie, X., & Yin, R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.13547 · Cited by 3
Why Is RLHF Alignment Shallow? A Gradient AnalysisPreprint
Young, R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.04851 · Cited by 3
Sirens' Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMsPreprint
Ling, Z., Hu, P., Gao, X., Ma, X., Zhou, M., Feng, J., Lu, S., Zhang, D., & Zhu, B. B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.13847 · Cited by 3
Don't Ask: Understanding What Numeric Self-Reports Measure in Language ModelsPreprint
Martorell, N., & Bianchi, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.18893 · Cited by 3
On the Failure of Topic-Matched Contrast Baselines in Multi-Directional Refusal AbliterationPreprint
Petrov, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.22061 · Cited by 4
Steering Code LLMs with Activation Directions for Language and Library ControlPreprint
Rahman, M. M., Guha, A., & Menon, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.23629 · Cited by 3
Saguer, I. L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.27412 · Cited by 3
Reasoning Models Know What's Important, and Encode It in Their ActivationsPreprint
Nikankin, Y., Tutek, M., Ashuach, T., Rosenfeld, J., & Belinkov, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.18307 · Cited by 3
Dynamic Adversarial Fine-Tuning Reorganizes Refusal GeometryPreprint
Lan, W., Li, S., Lai, X., Wu, M., Yang, J., Shen, H., & Yang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.27019 · Cited by 4
Sun, B., Li, C., Yang, Y., Wang, Y., & Xiao, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.27861 · Cited by 3
When No Benchmark Exists: Validating Comparative LLM Safety Scoring Without Ground-Truth LabelsPreprint
Gautam, S., Schwall, F., Olstad, A. W., Ruiz, F. V., Torpmann-Hagen, B., Bjørklund, S. M. S., Moonen, L., Pettersen, K. H., & Riegler, M. A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.06652 · Cited by 3
Tensor Product Representation Probes Reveal Shared Structure Across Linear DirectionsPreprint
Lee, A., Viégas, F. B., & Wattenberg, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.09967 · Cited by 3
Before the Last Token: Diagnosing Final-Token Safety Probe FailuresPreprint
Doda, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.12726 · Cited by 3
Better Accuracies, Worse Reasoning: A Step-Level Audit of Medical Chain-of-Thought DistillationPreprint
Jiang, Z., Peng, X., Teng, F., Fu, Z., Kim, Y., Mi, J., Li, Z., & Wu, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.28301 · Cited by 3
When Think-with-Image Meets Safety: What Determines Multimodal Jailbreak Robustness?Preprint
Tian, Y., Hu, B., Wu, F., Li, X., Lu, B., & Gong, N. Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.27932 · Cited by 3
Task-Focused Memorization for Multimodal AgentsPreprint
Zou, T., He, Y., Qiu, T., Lin, Y., & Li, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.31075 · Cited by 3
DataShield: Safety-degrading Data Filtering for LLM Benign Instruction Fine-TuningPreprint
Zhang, J., Zhou, Q., Deng, X., Jiang, W., Pan, J., & Zhu, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.00160 · Cited by 3
Subspace-Aware Sparse Autoencoders for Effective Mechanistic InterpretabilityPreprint
Dalili, S. A., & Mahdavi, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.06333 · Cited by 3
Cao, Q., Lou, J., Liu, M., Feng, W., Li, D., & Luu, A. T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.08682 · Cited by 3
Quantifying Subliminal Behavioral Transfer Ratios in Language Model DistillationPreprint
Konig, U., Kazmi, H., Li, R., & Chaudhary, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.11270 · Cited by 3
AnchorKV: Safety-Aware KV Cache Compression via Soft Penalty with a Refusal AnchorPreprint
Ni, N., & Lao, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.17872 · Cited by 3
Where Do Models Find Happiness? Emotion Vectors in Open-Source LLMsPreprint
van der Ben, S., Baur, R., Metz, Y., & El‐Assady, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.26987 · Cited by 3
Harnessing Textual Refusal Directions for Multimodal SafetyPreprint
Moreno D’Incà, Sebe, N., & Mancini, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.31876 · Cited by 4
Faithfulness to Refusal: A Causal Audit of Neuron SelectorsPreprint
Eswar, A., Seth, P., Avaiya, U., & Sankarapu, V. K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.05355 · Cited by 3
An Emergent Mirage: Is Emergent Misalignment and Realignment Indeed a Robust Phenomenon?Preprint
Rao, A., Gong, L., Hu, B., & Naik, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.09053 · Cited by 3
Prosvirnin, R., Minchenkov, V., Soldatov, A., Bashun, V., & Sergeev, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.12792 · Cited by 3
Emergent Misalignment Recruits a Pre-existing Persona SubspacePreprint
Nadaf, M. S. B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.21356 · Cited by 3
Routing Subspaces: Auditing Evaluation-to-Deployment Mismatch in Fine-Tuned Language ModelsPreprint
Konrad, P. M., Tanyel, T., & Ayvaz, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.20436 · Cited by 3
When Refusal Looks Safe: The Refusal-Cue Shortcut in Safety Guard ModelsPreprint
Feng, Y., Zang, C., Shen, C., Miao, R., Teng, G., Cai, W., & Ye, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.03201 · Cited by 3
CircuitSteer: Geometrically Aligned Multi-Layer Steering via Sparse Autoencoder CircuitsPreprint
Saadatinia, M., Razmara, P., Aryashad, A., Abbasi, A., & Azizi, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.05732 · Cited by 3
The Illusion of Cross-Lingual Safety in Low-Resource LanguagesPreprint
Oppong, A., Sahil, P. S., Belay, T. D., Mukhtar, M. I., Abdu, E. A., Abdullahi, T., Oparebea, J., Aliyu, S. M., Abdulmumin, I., Chilala, A. J., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.11146 · Cited by 3
Locating and Controlling Implicit Personalization in Large Language ModelsPreprint
Yan, Y., Wu, S., & Le, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.11735 · Cited by 3
Tunable Tool-Call Rates in LLM Agents via Representation SteeringPreprint
Chen, Y., Siu, V., Liu, Y., Song, D., & Wang, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.25198 · Cited by 3
Thomson: Continual Learning of Frontier Models for SovereignAIPreprint
Chen, S., Parker, J., Bang, Y., Bean, A., Seedat, N., Winzeck, S., Glazko, D., Zgraggen, J., Yu, F., Arnott, S., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.27147 · Cited by 3
Leong, J. W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.28502 · Cited by 3
How Language Models Choose Sides: Internal Representations of Instruction HierarchyPreprint
Balp-Straffon, E., Hsu, C.-H., Gadhvi, R., Dev, S., McDougall, C. S., & Mujumdar, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.28648 · Cited by 3
Safin-1: Safety from Within through Memory-Native State EvolutionPreprint
Zhang, M., Yang, K., Yu, S., Hua, E., Chen, Z., Jin, C., Zheng, J., Zhang, Y., Ma, Z., Zhou, J., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.00092 · Cited by 3
Wang, Y., Dang, J., Wang, B., Wu, Y., Zhang, W., Peng, H., Liu, J., Hu, B., Tian, Q., & Chua, T.-S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.22135 · Cited by 3
Li, H., Chen, M., Ren, W., Li, D., Tian, Z., Huang, J., & Wang, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.36484 · Cited by 3
Monitoring latent reasoning in looped language modelsPreprint
Olszowiec, P., Tomaszewska, P., Żelaszczyk, M., Byra, M., & Chung, N. C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.18127 · Cited by 3
Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
Lan, M., Torr, P., Meek, A., Khakzar, A., Krueger, D., & Barez, F. (2024) · arXiv · doi:10.48550/arxiv.2410.06981 · Cited by 2
EnSToM: Enhancing Dialogue Systems with Entropy-Scaled Steering Vectors for Topic Maintenance
Suh, H., Jeon, Y., Kang, D., Park, T., Min, Y., & Lee, G. K. K. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.1264 · Cited by 2
Navigating through the hidden embedding space: steering LLMs to improve mental health assessment
Ravenda, F., Bahrainian, S. A., Raballo, A., & Mira, A. (2026) · ACM Symposium on Applied Computing · doi:10.1145/3748522.3779887 · Cited by 2
Decomposed Trust: Privacy, Adversarial Robustness, Ethics, and Fairness in Low-Rank LLMs
Asante, D. A., Chowdhury, M. M., & Li, Y. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.93 · Cited by 2
CoT defender: Preemptive chain-of-thought occupation for jailbreak attack mitigation
Li, X., Liu, J., Tang, Y., Xie, Z., Wang, Y., Yu, X., Zhao, L., & Huang, B. (2026) · Neural Networks · doi:10.1016/j.neunet.2026.108574 · Cited by 2
Attributing and Exploiting Safety Vectors through Global Optimization in Large Language Models
Chu, F., Chen, J., Wang, Y., Wang, J., Fu, Z., Ji, S., Dong, L., & Li, S. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.378 · Cited by 2
Fluid Reasoning Representations
Kharlapenko, D., Zhang, T. J., Singh, A., Stolfo, A., Conmy, A., Sachan, M., & Jin, Z. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.659 · Cited by 2
Singh, H., Xu, Z., Subramanyam, A. V., & Kankanhalli, M. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1920 · Cited by 2
Ieong, L.-L., Chen, C.-C., Yang, C.-K., Huang, Y.-H., Cheng, A.-Y., & Lee, H.-Y. (2026) · Interspeech (USB) · doi:10.21437/interspeech.2026-554 · Cited by 2
Activation Surgery: Jailbreaking White-Box LLMs Without Touching the Prompt
Jenny, M., Dentan, J., Haddad-Vanier, S., & Krajecki, M. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-38702-8_4 · Cited by 2
AlignTree: Efficient Defense Against LLM Jailbreak Attacks
Goren, G., Katz, S., & Wolf, L. (2026) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v40i44.41074 · Cited by 2
FreakOut-LLM: The Effect of Emotional Stimuli on Safety Alignment
Kuznetsov, D., Cohen, O., Shistik, K., Shabtai, A., & Puzis, R. (2026) · doi:10.18653/2026.findings-aacl.198 · Cited by 3
Karnatak, N., Chatila, M., Hernandez, D. A. P., Yazdanfar, R., DUGAS, M. A., & Vakis, R. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791062 · Cited by 2
Safety Alignment Should be Made More than just A Few Attention Heads
Huang, C. J., Zhang, Z., Yue, J., Sheng, J., Li, Q., Zhang, C., & Liu, T. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11465040 · Cited by 2
The Age of Curiosity Meets the Age of AI: Benchmarking Child Safety in Large Language Models
Arif, S., Borah, A., & Mihalcea, R. F. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.781 · Cited by 2
Low-Resource Safety Failures Are Action Failures, Not Representation Failures
Aziz, R., Hanif, I. A., & Koto, F. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.287 · Cited by 2
HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models
Liu, S., Bo, Y., Zhao, Q., Huang, C., Chen, X., Liu, Y., & Du, M. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1095 · Cited by 2
Steering Vectors are an Adversarial Attack Surface
Aidakhmetov, A., Crisostomi, D., Mencattini, T., Minut, A. R., Masi, I., & Rodolà, E. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.2040 · Cited by 2
Skin-Deep: A Geometric Diagnostic for Alignment Fragility in Large Language Model Representations
Lee, D., Lee, J., Lee, S., Hong, S., Son, S., Eo, S., Seo, J., & Lim, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.22676 · Cited by 3
Symbolic Mechanistic Data Attribution: Tracing Training Influence to Learned Behavioral Policies
Habibi, R., Lee, D., & El-Nasr, M. S. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1671 · Cited by 2
The Geometry of Refusal: Linear Instability in Safety-Aligned LLMs
Ratnakar, S., & Vats, K. (2026) · Workshop on Trustworthy NLP (TrustNLP) · doi:10.18653/v1/2026.trustnlp-main.51 · Cited by 2
SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models
Zeng, X., Yang, X., Zhang, J., Yan, H., Chen, X., Wei, K., Kang, H., & Tian, Y. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.916 · Cited by 2
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
Zhang, H., Wang, D., Liu, Y., Chen, K., & Wang, W. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.260 · Cited by 2
Silicon Sampling via Cross-Survey Transfer
Ku, C.-T., Hsu, C., Huang, P.-C., Liu, F. C.-S., Cheng, I.-L., & Kang, Y. (2026) · IEEE International Conference on Information Reuse and Integration (IRI) · doi:10.1109/iri69576.2026.00070 · Cited by 2
Prosody-driven Jailbreaks in Audio LLMs: A Controlled Study and Mechanistic Analysis
Qian, J., & Li, J. (2026) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3767308.3835306 · Cited by 2
MMAligner: Safeguarding Multimodal Large Language Models through Representation Calibration
Zhang, S., Guo, K., Wang, Z., Li, X., Zhao, L., Hu, H., Shen, C., & Wang, Q. (2026) · ACM SIGSAC Conference on Computer and Communications Security (CCS) · doi:10.48550/arxiv.2608.05909 · Cited by 2
Metadata
- Authors
- Arditi, A., Obeso, O., Syed, A., Paleka, D., Panickssery, N., Gurnee, W., & Nanda, N.
- ORCID
- Gurnee, Wes
- Venue
- Advances in Neural Information Processing Systems 37
- Published
- 2024
- Pages
- 136037-136083
- ISSN
- 1049-5258 (ISSN-L)
- OpenAlex
- W4415800641
- Cited by
- 70
- References
- 0
- Citations by year
- 2026: 61 · 2025: 9
- FWCI
- 386.89
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- neural information processing systems (publishedVersion)
- Topics
- Social Sciences › Arts and Humanities › Language and Linguistics › Language, Discourse, Communication Strategies
- Social Sciences › Arts and Humanities › Language and Linguistics › Linguistics and Language Analysis
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Keywords
- refusal, language models, large language models
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.