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
Open Problems in Mechanistic Interpretability
Sharkey, L., Chughtai, B., Batson, J., Lindsey, J., Wu, J., Bushnaq, L., Goldowsky-Dill, N., Heimersheim, S., Ortega, A., Bloom, J., et al. (2025) · arXiv · doi:10.48550/arxiv.2501.16496 · Cited by 291
Self-citationA Practical Review of Mechanistic Interpretability for Transformer-Based Language ModelsPreprint
Rai, D., Zhou, Y., Feng, S., Saparov, A., & Yao, Z. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.02646 · Cited by 296
A Primer on the Inner Workings of Transformer-based Language ModelsPreprint
Ferrando, J., Sarti, G., Bisazza, A., & Costa‐jussà, M. R. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2405.00208 · Cited by 218
A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and DeploymentPreprint
Wang, K., Zhang, G., Zhou, Z., Wu, J., Yu, M., Zhao, S., Yin, C., Fu, J., Yan, Y., Luo, H., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.15585 · Cited by 196
Representation Noising: A Defence Mechanism Against Harmful Finetuning
Rosati, D., Wehner, J., Williams, K., Bartoszcze, Ł., Atanasov, D., Gonzales, R., Majumdar, S., Maple, C. R., Sajjad, H., & Rudzicz, F. (2024) · neural information processing systems · doi:10.52202/079017-0402 · Cited by 173
Safety Layers in Aligned Large Language Models: The Key to LLM Security
Li, S., Yao, L., Zhang, L., & Li, Y. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2408.17003 · Cited by 170
Towards Interpreting Visual Information Processing in Vision-Language Models
Neo, C., Ong, C.-H. L., Torr, P. H. S., Geva, M., Krueger, D., & Barez, F. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2410.07149 · Cited by 166
Usable XAI: 10 Strategies Towards Exploiting Explainability in the LLM Era
Wu, X., Zhao, H., Zhu, Y., Shi, Y., Yang, F., Hu, L., Liu, T., Zhaı, X., Yao, W., Li, J., et al. (2026) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3816150 · Cited by 150
Programming Refusal with Conditional Activation Steering
Lee, B. W., Padhi, I., Ramamurthy, K. N., Miehling, E., Dognin, P., Nagireddy, M., & Dhurandhar, A. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2409.05907 · Cited by 149
The Assistant Axis: Situating and Stabilizing the Default Persona of Language ModelsPreprint
Lu, C., Gallagher, J., Michala, J., Fish, K., & Lindsey, J. W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.10387 · Cited by 151
LLMs Encode Harmfulness and Refusal Separately
Zhao, J., Huang, J., Wu, Z., Bau, D., & Shi, W. (2025) · neural information processing systems · doi:10.52202/085713-4688 · Cited by 130
Benchmarking Distributional Alignment of Large Language Models
Meister, N., Guestrin, C., & Hashimoto, T. (2025) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2025.naacl-long.2 · Cited by 116
Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models
Ferrando, J., Obeso, O., Rajamanoharan, S., & Nanda, N. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2411.14257 · Cited by 115
Self-citationTowards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons
Chen, J., Wang, X., Yao, Z., Bai, Y., Hou, L., & Li, J. (2025) · neural information processing systems · doi:10.52202/085713-0426 · Cited by 113
Understanding Reasoning in Thinking Language Models via Steering VectorsPreprint
Venhoff, C., Arcuschin, I., Torr, P., Conmy, A., & Nanda, N. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.18167 · Cited by 106
Self-citationWhy AI Is WEIRD and Shouldn't Be This Way: Towards AI for Everyone, with Everyone, by Everyone
Mihalcea, R. F., Ignat, O., Bai, L., Borah, A., Chiruzzo, L., Jin, Z., Kwizera, C., Nwatu, J. C., Poria, S., & Solorio, T. (2025) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v39i27.35092 · Cited by 105
The Geometry of Refusal in Large Language Models: Concept Cones and Representational IndependencePreprint
Wollschläger, T., Elstner, J., Geisler, S., Cohen-Addad, V., Günnemann, S., & Gasteiger, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.17420 · Cited by 104
Saphra, N., & Wiegreffe, S. (2024) · BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP) · doi:10.18653/v1/2024.blackboxnlp-1.30 · Cited by 88
A Comprehensive Survey of Machine Unlearning Techniques for Large Language ModelsPreprint
Geng, J., Li, Q., Woisetschläger, H., Chen, Z., Cai, F., Wang, Y., Nakov, P., Jacobsen, H., & Karray, F. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.01854 · Cited by 84
BEEAR: Embedding-based Adversarial Removal of Safety Backdoors in Instruction-tuned Language Models
Zeng, Y., Sun, W., Huynh, T., Song, D. X., Li, B., & Jia, R. (2024) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2024.emnlp-main.732 · Cited by 85
Automatic Pseudo-Harmful Prompt Generation for Evaluating False Refusals in Large Language ModelsPreprint
An, B., Zhu, S., Zhang, R., Panaitescu-Liess, M.-A., Xu, Y., & Huang, F. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2409.00598 · Cited by 79
Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks
Wang, H., Gang, W., & Zhang, H. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.02787 · Cited by 76
What Makes and Breaks Safety Fine-tuning? A Mechanistic Study
Jain, S., Lubana, E. S., Öksüz, K., Joy, T., Torr, P. H. S., Sanyal, A., & Dokania, P. K. (2024) · neural information processing systems · doi:10.52202/079017-2963 · Cited by 73
Operationalizing a Threat Model for Red-Teaming Large Language Models (LLMs)
Verma, A., Krishna, S., Gehrmann, S., Seshadri, M., Pradhan, A. R., Ault, T., Barrett, L., Rabinowitz, D., Doucette, J. A., & Phan, N. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2407.14937 · Cited by 70
Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilities
Che, Z., Casper, S., Kirk, R., Satheesh, A., Slocum, S., McKinney, L. E., Gandikota, R., Ewart, A., Rosati, D., Wu, Z., et al. (2025) · Transactions on machine learning research · doi:10.48550/arxiv.2502.05209 · Cited by 71
Zhang, Z., Xu, W., Wu, F., & Reddy, C. K. A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.08054 · Cited by 69
Steering Without Side Effects: Improving Post-Deployment Control of Language ModelsPreprint
Stickland, A. C., Lyzhov, A., Pfau, J., Mahdi, S., & Bowman, S. R. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2406.15518 · Cited by 66
A Survey of Test-time Compute: From Intuitive Inference to Deliberate Reasoning
Ji, Y., Li, J., Xiang, Y., Ye, H., Wu, K., Kai, Y., Xu, J., Mo, L., & Zhang, M. (2026) · Computational Linguistics · doi:10.1162/coli.a.629 · Cited by 66
On the Role of Attention Heads in Large Language Model Safety
Zhou, Z., Yu, H., Zhang, X., Xu, R., Huang, F., Wang, K., Liu, Y., Fang, J., & Li, Y. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2410.13708 · Cited by 59
AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint
Sheng, L., Shen, C. P., Zhao, W., Fang, J., Liu, X., Liang, Z., Wang, X., Zhang, A., & Chua, T. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2506.07022 · Cited by 57
Schoene, A. M., & Canca, C. (2025) · IEEE International Symposium on Technology and Society (ISTAS) · doi:10.1109/istas65609.2025.11269647 · Cited by 57
LLM Unlearning via Neural Activation Redirection
Shen, W. F., Qiu, X., Kurmanji, M., Iacob, A., Sani, L., Chen, Y., Cancedda, N., & Lane, N. D. (2025) · neural information processing systems · doi:10.52202/085713-1475 · Cited by 56
Intrinsic Test of Unlearning Using Parametric Knowledge Traces
Hong, Y., Yu, L., Yang, H., Ravfogel, S., & Geva, M. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.985 · Cited by 55
On Evaluating the Durability of Safeguards for Open-Weight LLMs
Qi, X., Wei, B., Carlini, N., Huang, Y., Xie, T., He, L., Jagielski, M., Nasr, M., Mittal, P., & Henderson, P. (2024) · International Conference on Learning Representations · doi:10.48550/arxiv.2412.07097 · Cited by 57
Obfuscated Activations Bypass LLM Latent-Space Defenses
Bailey, L., Serrano, A., Sheshadri, A., Seleznyov, M., Taylor, J., Jenner, E., Hilton, J., Casper, S., Guestrin, C., & Emmons, S. (2024) · International Conference on Learning Representations · doi:10.48550/arxiv.2412.09565 · Cited by 53
From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit
COSTA, V. O. D., Fel, T., Lubana, E. S., Tolooshams, B., & Ba, D. E. (2025) · neural information processing systems · doi:10.52202/085713-1154 · Cited by 52
Persistent Pre-Training Poisoning of LLMs
Zhang, Y., Rando, J., Evtimov, I., Chi, J., Smith, E. M., Carlini, N., Tramèr, F., & Ippolito, D. (2024) · International Conference on Learning Representations · doi:10.48550/arxiv.2410.13722 · Cited by 53
Understanding (Un)Reliability of Steering Vectors in Language ModelsPreprint
Braun, J., Eickhoff, C., Krueger, D., Bahrainian, S. A., & Krasheninnikov, D. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.22637 · Cited by 52
Don't Overthink It: A Survey of Efficient R1-style Large Reasoning ModelsPreprint
Yue, L., Du, Y., Wang, Y., Gao, W., Yao, F., Li, W., Liu, Y., Xu, Z., Liu, Q., Di, S., et al. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.02120 · Cited by 53
AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender
Zhao, W., Guo, J., Hu, Y., Deng, Y., Zhang, A., Sui, X., Han, X., Zhao, Y., Qin, B., Chua, T., et al. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.1248 · Cited by 52
Refusal Direction is Universal Across Safety-Aligned Languages
Wang, X., Wang, M., Liu, Y., Schuetze, H., & Plank, B. (2025) · neural information processing systems · doi:10.52202/085713-1086 · Cited by 50
SafeSwitch: Steering Unsafe LLM Behavior via Internal Activation Signals
Han, P., Qian, C., Chen, X., Zhang, Y., Ji, H., & Zhang, D. (2025) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2025.findings-emnlp.366 · Cited by 49
Understanding Jailbreak Success: A Study of Latent Space Dynamics in Large Language Models
Ball, S., Kreuter, F., & Panickssery, N. (2026) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2026.eacl-long.12 · Cited by 49
Self-citationLanguage Model Alignment in Multilingual Trolley Problems
Jin, Z., Kleiman-Weiner, M., Piatti, G., Levine, S., Liu, J., Gonzalez, F., Ortu, F., Strausz, A., Sachan, M., Mihalcea, R. F., et al. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2407.02273 · Cited by 48
Towards Understanding the Fragility of Multilingual LLMs against Fine-Tuning Attacks
Poppi, S., Yong, Z. X., He, Y., Chern, B., Zhao, H., Yang, A., & Chi, J. (2025) · Findings of the Association for Computational Linguistics: NAACL · doi:10.18653/v1/2025.findings-naacl.126 · Cited by 48
ICLR: In-Context Learning of Representations
Park, C. F., Lee, A., Lubana, E. S., Yang, Y., Okawa, M., Nishi, K., WATTENBERG, M. P., & Tanaka, H. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2501.00070 · Cited by 47
Mitigating Overthinking in Large Reasoning Models via Manifold Steering
Huang, Y., Chen, H., Ruan, S., Zhang, Y., Wei, X., & Dong, Y. (2025) · neural information processing systems · doi:10.52202/085713-3430 · Cited by 46
FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering
Li, Y., Fan, Z., Chen, R., Gai, X., Gong, L., Zhang, Y., & Liu, Z. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.589 · Cited by 45
Refusal in LLMs is an Affine FunctionPreprint
Märshall, T., Scherlis, A., & Belrose, N. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2411.09003 · Cited by 46
Probing and Steering Evaluation Awareness of Language ModelsPreprint
Nguyen, J., Hoang, K., Attubato, C. L., & Hofstätter, F. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.01786 · Cited by 46
Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal ActivationsPreprint
Ji-An, L., 熊华东, Wilson, R. C., Mattar, M. G., & Benna, M. K. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.13763 · Cited by 45
Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoders
Deng, B., Wan, Y., Yang, B., Zhang, Y., & Feng, F. (2025) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2025.acl-long.229 · Cited by 44
The Hidden Dimensions of LLM Alignment: A Multi-Dimensional Analysis of Orthogonal Safety DirectionsPreprint
Pan, W., Liu, Z., Chen, Q., Zhou, X., Yu, H., & Jia, X. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.09674 · Cited by 45
Wang, X., Hu, C., Röttger, P., & Plank, B. (2024) · International Conference on Learning Representations · doi:10.48550/arxiv.2410.03415 · Cited by 43
Manifold Steering Reveals the Shared Geometry of Neural Network Representation and BehaviorPreprint
Wurgaft, D., Rager, C., Kowal, M., Shyam, V., Feucht, S., Bhalla, U., Haklay, T., Bigelow, E., Sarfati, R., McGrath, T., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.05115 · Cited by 44
Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space
Park, C. F., Okawa, M., Lee, A., Tanaka, H., & Lubana, E. S. (2024) · neural information processing systems · doi:10.52202/079017-2689 · Cited by 42
Controlling Large Language Model Agents with Entropic Activation SteeringPreprint
Rahn, N., D’Oro, P., & Bellemare, M. G. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2406.00244 · Cited by 42
Interpreting Physics in Video World ModelsPreprint
Joseph, S., Garrido, Q., Balestriero, R., Kowal, M. D., Fel, T., Bakhtiari, S., Richards, B., & Rabbat, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.07050 · Cited by 43
Qwen-Scope: Turning Sparse Features into Development Tools for Large Language ModelsPreprint
Deng, B., Wang, X., Wang, Y., Wan, Y., Ma, Y., Yang, B., Wei, H., Tang, J., Lin, H., Gao, R., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.11887 · Cited by 43
Preference Tuning For Toxicity Mitigation Generalizes Across Languages
Li, X., Yong, Z., & Bach, S. H. (2024) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2024.findings-emnlp.784 · Cited by 41
Zhang, H., Zhang, Z., Wang, M., Su, Z., Wang, Y., Wang, Q., Yuan, S., Nie, E., Duan, X., Xue, Q., et al. (2026) · Computer Science Review · doi:10.1016/j.cosrev.2026.101011 · Cited by 41
Towards Inference-time Category-wise Safety Steering for Large Language ModelsPreprint
Bhattacharjee, A., Ghosh, S., Rebedea, T. E., & Parisien, C. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.01174 · Cited by 40
Calibrating Verbal Uncertainty as a Linear Feature to Reduce HallucinationsPreprint
Ji, Z., Yu, L., Koishekenov, Y., Bang, Y., Hartshorn, A., Schelten, A., Zhang, C., Fung, P., & Cancedda, N. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.14477 · Cited by 40
SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Models
He, Z., Jin, M., Shen, B., Payani, A., Zhang, Y., & Du, M. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.112 · Cited by 38
One-shot Optimized Steering Vectors Mediate Safety-relevant Behaviors in LLMsPreprint
Dunefsky, J., & Cohan, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.18862 · Cited by 38
Hazra, R., Layek, S., Banerjee, S., & Poria, S. (2024) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2024.emnlp-main.1212 · Cited by 37
Sparse Autoencoder Features for Classifications and Transferability
Gallifant, J., Chen, S., Sasse, K., Aerts, H. J. W. L., Hartvigsen, T., & Bitterman, D. S. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.1521 · Cited by 38
He, Z., Zhao, H., Qiao, Y., Yang, F., Payani, A., Ma, J., & Du, M. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37676-3_11 · Cited by 38
FLRT: Fluent Student-Teacher RedteamingPreprint
Thompson, T., & Sklar, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.17447 · Cited by 35
Reasoning-Finetuning Repurposes Latent Representations in Base ModelsPreprint
Ward, J., Lin, C., Venhoff, C., & Nanda, N. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.12638 · Cited by 35
Self-citationHou, A. B., Du, H., Wang, Y., Zhang, J., Wang, Z., Liang, P. P., Khashabi, D., Gardner, L., & He, T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.09639 · Cited by 35
Self-Destructive Language Model
Wang, Y., Zhu, R., & Wang, T. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2505.12186 · Cited by 35
CASE-Bench: Context-Aware SafEty Benchmark for Large Language ModelsPreprint
Sun, G., 战, 晓., Feng, S., Woodland, P. C., & Such, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2501.14940 · Cited by 34
LLM-Safety Evaluations Lack RobustnessPreprint
Beyer, T., Xhonneux, S., Geisler, S., Gidel, G., Schwinn, L., & Günnemann, S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.02574 · Cited by 36
Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation SteeringPreprint
Bigelow, E., Wurgaft, D., Wang, Y., Goodman, N. D., Ullman, T. D., Tanaka, H., & Lubana, E. S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2511.00617 · Cited by 35
Base Models Know How to Reason, Thinking Models Learn When
Venhoff, C., Arcuschin, I., Torr, P., Conmy, A., & Nanda, N. (2025) · Oxford University Research Archive (ORA) (University of Oxford) · doi:10.48550/arxiv.2510.07364 · Cited by 33
Self-citationTransferring Linear Features Across Language Models With Model Stitching
Chen, A., Merullo, J., Stolfo, A., & Pavlick, E. (2025) · neural information processing systems · doi:10.52202/085713-1621 · Cited by 33
Do BERT-Like Bidirectional Models Still Perform Better on Text Classification in the Era of LLMs?
Zhang, J., Huang, Y., Liu, S., Gao, Y., & Hu, X. (2025) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2025.findings-emnlp.1033 · Cited by 32
Position: Adversarial ML for LLMs Is Not Making Any ProgressPreprint
Rando, J., Zhang, J., Carlini, N., & Tramèr, F. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.02260 · Cited by 33
A Framework for Formalizing LLM Agent SecurityPreprint
Siu, V., He, J., Montgomery, K., Wang, Z., Gong, N. Z., Wang, C., & Song, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.19469 · Cited by 33
Training large language models on narrow tasks can lead to broad misalignment
Betley, J., Warncke, N., Sztyber-Betley, A., Tan, D., Bao, X., Soto, M., Srivastava, M., Labenz, N., & Evans, O. (2026) · Nature · doi:10.1038/s41586-025-09937-5 · Cited by 31
Balancing Large Language Model Alignment and Algorithmic Fidelity in Social Science Research
Lyman, A., Hepner, B., Argyle, L. P., Busby, E. C., Gubler, J. R., & Wingate, D. (2025) · Sociological Methods & Research · doi:10.1177/00491241251342008 · Cited by 32
Representation Bending for Large Language Model Safety
Yousefpour, A., Kim, T., Kwon, R. S., Lee, S., Jeung, W., Han, S., Wan, A., Ngan, H., Yu, Y., & Choi, J. (2025) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2025.acl-long.1173 · Cited by 31
Mechanistic indicators of understanding in large language models
Beckmann, P., & Queloz, M. (2026) · Philosophical Studies · doi:10.1007/s11098-026-02513-1 · Cited by 33
Djuhera, A., Kadhe, S., Ahmed, F., Zawad, S., & Boche, H. (2026) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2026.findings-acl.1761 · Cited by 32
Locking Down the Finetuned LLMs SafetyPreprint
Zhu, M., Yang, L., Wei, Y., Zhang, N., & Zhang, Y. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.10343 · Cited by 32
Extracting Unlearned Information from LLMs with Activation SteeringPreprint
Seyitoğlu, A., Kuvshinov, A., Schwinn, L., & Günnemann, S. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2411.02631 · Cited by 31
PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flipsPreprint
Coalson, Z., Woo, J., Lin, C. S., Qu, J., Sun, Y., Chen, S., Yang, L., Saileshwar, G., Nair, P., Fang, B., et al. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2412.07192 · Cited by 32
Jailbreaking to JailbreakPreprint
Kritz, J., Robinson, V., Vacareanu, R., Varjavand, B., Choi, M., Gogov, B., Scale Red Team, Yue, S., Primack, W. E., & Wang, Z. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.09638 · Cited by 30
SafeScientist: Toward Risk-Aware Scientific Discoveries by LLM AgentsPreprint
Zhu, K., Zhang, J., Qi, Z., Shang, N., Liu, Z., Han, P., Su, Y., Yu, H., & You, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.23559 · Cited by 30
RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors?Preprint
Gupta, R., & Jenner, E. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.14261 · Cited by 30
The Rogue Scalpel: Activation Steering Compromises LLM SafetyPreprint
Korznikov, A., Galichin, A. V., Dontsov, A., Rogov, O. Y., Oseledets, I., & Tutubalina, E. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.22067 · Cited by 31
MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent SkillsPreprint
Guo, W., Zeng, W., Liu, C., Jia, X., Xu, Y., Tang, L., Fang, Y., & Liu, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.07131 · Cited by 30
Tsui, K. (2025) · arXiv · doi:10.48550/arxiv.2507.02778 · Cited by 29
Understanding Refusal in Language Models with Sparse Autoencoders
Yeo, W. J., Prakash, N., Neo, C., Satapathy, R., Lee, R. K.-W., & Cambria, E. (2025) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2025.findings-emnlp.338 · Cited by 30
Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinations
Ji, Z., Lei, Y., Koishekenov, Y., Bang, Y., Hartshorn, A., Schelten, A., Zhang, C., Fung, P., & Cancedda, N. (2025) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2025.emnlp-main.187 · Cited by 29
ASIDE: Architectural Separation of Instructions and Data in Language Models
Zverev, E., Kortukov, E., Panfilov, A., Volkova, A., Tabesh, S., Lapuschkin, S., Samek, W., & Lampert, C. H. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2503.10566 · Cited by 30
Subliminal Learning Is Steering Vector DistillationPreprint
Blank, C., Bhatia, A., Rajamanoharan, S., Conmy, A., & Nanda, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.00995 · Cited by 29
Self-citationThe Reasoning-Memorization Interplay in Language Models Is Mediated by a Single Direction
Hong, Y., Cao, M., Zhou, D., Yu, L., & Jin, Z. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.1111 · Cited by 28
PII-Scope: A Comprehensive Study on Training Data PII Extraction Attacks in LLMsPreprint
Nakka, K. K., Frikha, A., Mendes, R., Jiang, X., & Zhou, X. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.06704 · Cited by 29
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.