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 401–500 of 1,000
Ma, X.-X., Gao, K.-F., Lu, Z.-Y., Jiang, W., Fang, H., Wu, H., Chen, B., & Xia, S.-T. (2026) · Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2026.acl-long.1123 · Cited by 4
Rai, D., Miller, S., Moran, K., & Yao, Z. (2025) · neural information processing systems · doi:10.52202/085713-4418 · Cited by 5
On Mechanistic Circuits for Extractive Question-AnsweringPreprint
Basu, S., Morariu, V. I., Wang, Z., Rossi, R. A., Zhao, C., Feizi, S., & Manjunatha, V. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.08059 · Cited by 5
Ravindran, S. K. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.09406 · Cited by 5
On the Limitations of Steering in Language Model AlignmentPreprint
Chebrolu, N., Jaidka, K., & Yeo, G. C. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.01162 · Cited by 5
When Do LLMs Admit Their Mistakes? Understanding The Role Of Model Belief In RetractionPreprint
Yang, Y., & Jia, R. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.16170 · Cited by 5
From Directions to Cones: Exploring Multidimensional Representations of Propositional Facts in LLMsPreprint
Yu, K. S., Bulusu, V., Yasunaga, O., Lau, C., Blondin, C., O’Brien, S., Zhu, K., & Sharma, V. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.21800 · Cited by 5
Halloran, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.23634 · Cited by 5
Wu, S., Jin, G., Huang, W., Wang, J., & Huang, X. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.00373 · Cited by 5
Latent Adversarial Training Improves the Representation of RefusalPreprint
Abbas, A., Petrova, N., Lyons, H. A., & Pérez-Campanero, N. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.18872 · Cited by 5
LatentBreak: Jailbreaking Large Language Models through Latent Space FeedbackPreprint
Mura, R., Piras, G., Lukošiūtė, K., Pintor, M., Karbasi, A., & Biggio, B. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.08604 · Cited by 5
InverseScope: Scalable Activation Inversion for Interpreting Large Language ModelsPreprint
Luo, Y., Zhou, Z., & Dong, B. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.07406 · Cited by 5
Genre Controlled Music Generation via Activation SteeringPreprint
Narashiman, S., Mathur, P., Panda, D., Joe, J. K., R, H. M., Veerakumar, A., Krishna, A., & Keerthiharan, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.10225 · Cited by 5
Sysformer: Safeguarding Frozen Large Language Models with Adaptive System Prompts
Sharma, K., Jin, Y., Rakesh, V., Dou, Y., Pan, M., Das, M., & Kumar, S. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2506.15751 · Cited by 5
From Concepts to Components: Concept-Agnostic Attention Module Discovery in Transformers
Su, J., Kempe, J., & Ullrich, K. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2506.17052 · Cited by 5
Deceptive Automated Interpretability: Language Models Coordinating to Fool Oversight SystemsPreprint
Lermen, S., Dziemian, M., & Antolín, N. P.-C. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.07831 · Cited by 6
The Blessing and Curse of Dimensionality in Safety AlignmentPreprint
Teo, R. S. Y., Abdullaev, L. U., & Nguyen, T. M. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.20333 · Cited by 5
Steering When Necessary: Flexible Steering Large Language Models with BacktrackingPreprint
Cheng, Z., Gan, J., Jiang, Z., Wang, C., Yin, Y., Luo, X., Fu, Y., & Gu, Q. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.17621 · Cited by 6
The LLM Has Left The Chat: Evidence of Bail Preferences in Large Language ModelsPreprint
Ensign, D., Sleight, H., & Fish, K. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2509.04781 · Cited by 5
Sequences of Logits Reveal the Low Rank Structure of Language Models
Golowich, N., Liu, A., & Shetty, A. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2510.24966 · Cited by 5
Maar, J., Paperno, D., McDougall, C. S., & Nanda, N. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2601.20164 · Cited by 5
Self-citationS3-CoT: Self-Sampled Succinct Reasoning Enables Efficient Chain-of-Thought LLMsPreprint
Du, Y., Zhao, S., Gao, Y., Zhao, D., Lin, Q., Ma, M., Li, J., Jiang, Y., He, K., Xu, Q., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.01982 · Cited by 5
BLOCK-EM: Preventing Emergent Misalignment via Latent BlockingPreprint
Ustaomeroglu, M., & Qu, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.00767 · Cited by 5
Steering to Say No: Configurable Refusal via Activation Steering in Vision Language ModelsPreprint
Yang, J., Liu, S., Yang, Y., & Lee, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.07013 · Cited by 6
Steer2Edit: From Activation Steering to Component-Level EditingPreprint
Sun, C.-E., Yan, G., Wang, Z., & Weng, T.-W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.09870 · Cited by 5
Orgad, H., Wei, B., Zheng, K., WATTENBERG, M. P., Henderson, P., Goldfarb-Tarrant, S., & Belinkov, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.09544 · Cited by 5
SafeDream: Safety World Model for Proactive Early Jailbreak DetectionPreprint
Yan, B., Lin, W., Zhu, Y., & Wang, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.16824 · Cited by 5
Harmful Intent as a Geometrically Recoverable Feature of LLM Residual StreamsPreprint
Saguer, I. L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.18901 · Cited by 5
FLIPS: Instance-Fingerprinting for LLMs via Pseudo-random SequencesPreprint
Richardeau, G., Dashyan, G., Merrer, E. L., & Trédan, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.03330 · Cited by 5
On the Inseparability of Instructions and Data in Shared-Embedding Sequence ModelsPreprint
Pant, D., Lohani, S., & Kumar, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.27567 · Cited by 5
Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge BiasPreprint
Xu, Z., Li, S., Liu, H., Wang, X., Li, S., Song, Z., & Chen, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.11871 · Cited by 5
DecodeShare: Tracing the Shared Subspace of LLM Decode-Time DecisionsPreprint
Shao, Z., Zhang, L., Cui, K., Wang, Y., Jiang, T., Ye, H., Wang, Q., Du, Z., Fu, Y., Yang, F., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.20469 · Cited by 5
Quantification of Large Language Model Distillation
Lee, S., Zhou, J., Ao, C., Li, K., Du, X., He, S., Wu, H., Liu, T., Liu, J., Alinejad-Rokny, H., et al. (2025) · Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2025.acl-long.248 · Cited by 4
Peng, G., Li, D. C. S., Luo, W., & Wang, H. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.1090 · Cited by 4
The Structural Safety Generalization Problem
Broomfield, J., Gibbs, T., Ingebretsen, G., Kosak-Hine, E., Nasir, T., Zhang, J., Iranmanesh, R., Pieri, S., Rabbany, R., & Pelrine, K. (2025) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2025.findings-acl.1142 · Cited by 4
Boosting Adversarial Transferability of Vision Transformers
Wang, Y., Zhang, C., Zhou, H., Ying, Z., Xiong, Z., Zhou, W., & Zhu, L. (2025) · IEEE Transactions on Dependable and Secure Computing · doi:10.1109/tdsc.2025.3605453 · Cited by 4
Better World Models Can Lead to Better Post-Training Performance
Gupta, P., Conklin, H., Leslie, S.-J., & Lee, A. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1012 · Cited by 4
Quantifying Self-Preservation Bias in Large Language Models
Migliarini, M., Pizzini, J. P., Moresca, L., Santini, V., Spinelli, I., & Galasso, F. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1238 · Cited by 4
From AI-Generated Content to Agentic Action: Security and Safety Threats in Generative AI
Zhang, Z.-L., Li, Q., Cao, J., Liu, L.-S., & Ni, J.-B. (2026) · Journal of Information and Intelligence · doi:10.1016/j.jiixd.2026.05.002 · Cited by 1
Out of Sight, Not Out of Mind: Unveiling Latent Attack in Latent-based Multi-Agent Systems
Wang, C., Huang, R., Sun, J., Wei, L., & Wu, Y. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.2093 · Cited by 5
Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate
Yi, J. S. K., Mueller, A., & Lee, D. (2026) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2026.acl-long.709 · Cited by 4
Where Steering Signals Come From: Activation Source Selection in Activation Steering
Ye, J., Ran, L., Yao, Z., Wang, C., Jiang, Y., Hou, L., Li, J., & Pan, L. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1829 · Cited by 4
Ma, H., Zou, H. P., Li, C., Ma, E., Su, Y., & Yu, P. S. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1208 · Cited by 4
Knowledge Graph Guided Evaluation of Abstention Techniques
Vasisht, K., Kaur, N., & Pruthi, D. (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.353 · Cited by 4
Steering Large Language Models for Machine Translation Personalization
Scalena, D., Sarti, G., Bisazza, A., Fersini, E., & Nissim, M. (2026) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.18653/v1/2026.eacl-long.217 · Cited by 4
Rethinking Explainable Machine Learning as Applied StatisticsPreprint
Bordt, S., Raidl, E., & von Luxburg, U. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2402.02870 · Cited by 4
GPT-2 Through the Lens of Vector Symbolic ArchitecturesPreprint
Knittel, J., Gangavarapu, T., Strobelt, H., & Pfister, H. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2412.07947 · Cited by 4
Does Editing Provide Evidence for Localization?Preprint
Wang, Z., & Veitch, V. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.11447 · Cited by 4
Robustly Identifying Concepts Introduced During Chat Fine-Tuning Using CrosscodersPreprint
Minder, J., & Dumas, C. (2025) · Qeios · doi:10.32388/r3sz5u · Cited by 4
Layered Unlearning for Adversarial RelearningPreprint
Qian, T., Suriyakumar, V. M., Wilson, A., & Hadfield-Menell, D. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2505.09500 · Cited by 4
Jailbreak Transferability Emerges from Shared Representations
Angell, R., Brinkmann, J., & He, H. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2506.12913 · Cited by 4
Feature-Aware Malicious Output Detection and MitigationPreprint
Dong, W., Li, P., Tian, Y., Zeng, X., Li, F., & Wang, S. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2504.09191 · Cited by 4
GRAF: Multi-turn Jailbreaking via Global Refinement and Active FabricationPreprint
Tang, H., Yan, L., Zhao, Y., Wang, S., Huang, J., & Yin, D. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2506.17881 · Cited by 4
RepV: Safety-Separable Latent Spaces for Scalable Neurosymbolic Plan VerificationPreprint
Yang, Y., Bhatt, N. P., Samineni, P., Siva, R., Wang, Z., & Topcu, U. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.26935 · Cited by 4
Any-Depth Alignment: Unlocking Innate Safety Alignment of LLMs to Any-Depth
Zhang, J., Estornell, A., Baek, D. D., Li, B., & Xu, X. (2025) · International Conference on Learning Representations · doi:10.48550/arxiv.2510.18081 · Cited by 4
Linearly Decoding Refused Knowledge in Aligned Language ModelsPreprint
Shrivastava, A., & Holtzman, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2507.00239 · Cited by 4
VISOR: Visual Input-based Steering for Output Redirection in Vision-Language ModelsPreprint
Phute, M., & Balakrishnan, R. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.08521 · Cited by 4
How Semantically Stable Are LLM Refusals? Measuring Confusion in Local Safety BoundariesPreprint
Anonto, R. A., Nahiyan, M. L. A., & Hassan, M. T. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.01037 · Cited by 4
Securing External Deeper-than-black-box GPAI EvaluationsPreprint
Tlaie, A., & Farrell, J. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.07496 · Cited by 4
How Effective Is Constitutional AI in Small LLMs? A Study on DeepSeek-R1 and Its PeersPreprint
Menke, A.-G. C., & Tan, P. X. (2025) · arXiv (Cornell University) · doi:10.13140/rg.2.2.28637.73442 · Cited by 4
Unveiling the Latent Directions of Reflection in Large Language ModelsPreprint
Chang, F.-C., Lee, Y.-T., & Wu, P.-Y. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2508.16989 · Cited by 4
Graph-Regularized Sparse Autoencoders for LLM Safety SteeringPreprint
Yeon, J., Cinus, F., Wu, Y., & Luceri, L. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.06655 · Cited by 4
Cristofano, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.16034 · Cited by 4
LLMs Can Unlearn Refusal with Only 1,000 Benign SamplesPreprint
Guo, Y., Xu, Z., Liu, S., Zheng, Z., & Kankanhalli, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.19231 · Cited by 4
Decomposing multimodal embedding spaces with group-sparse autoencoders
Kaushik, C., Barch, D. R., & Fanelli, A. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2601.20028 · Cited by 4
$C$-$ΔΘ$: Circuit-Restricted Weight Arithmetic for Selective RefusalPreprint
Kasliwal, A., Seth, P., & Sankarapu, V. K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.04521 · Cited by 4
Position: Capability Control Should be a Separate Goal From AlignmentPreprint
Siddiqui, S. A., Triantafillou, E., Krueger, D., & Weller, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.05164 · Cited by 4
Naseem, U. (2026) · doi:10.36227/techrxiv.177031352.23132160/v1 · Cited by 4
When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential ProcessingPreprint
Dadfar, Z. P. (2026) · arXiv (Cornell University) · doi:10.5281/zenodo.18567445 · Cited by 4
Fail-Closed Alignment for Large Language ModelsPreprint
Coalson, Z., Sohler, B., Gabriel, A., & Hong, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2602.16977 · Cited by 5
What Is the Alignment Tax?Preprint
Young, R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.00047 · Cited by 4
COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics
Sharma, K., & Trivedi, R. (2026) · International Conference on Learning Representations · doi:10.48550/arxiv.2603.06495 · Cited by 4
Jailbreak Scaling Laws for Large Language Models: Polynomial-Exponential CrossoverPreprint
Halder, I., Banerjee, A., & Pehlevan, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.11331 · Cited by 4
How Alignment Routes: Localizing, Scaling, and Controlling Policy Circuits in Language ModelsPreprint
Frank, G. N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.04385 · Cited by 4
Steering the Verifiability of Multimodal AI HallucinationsPreprint
Pang, J., Cheng, R., Ye, Z., Ma, X., Wu, Z., Huang, X., & Jiang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.06714 · Cited by 4
Linear Representations of Hierarchical Concepts in Language ModelsPreprint
SAKATA, M., Heinzerling, B., Ito, T., Yokoi, S., & Inui, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.07886 · Cited by 4
Calibrate-Then-Delegate: Safety Monitoring with Risk and Budget Guarantees via Model CascadesPreprint
Pona, E., Kazemi, M., Hosseini, M., Du, Y., Watson, D., Simeone, O., & Paoletti, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.14251 · Cited by 4
Surgical Repair of Insecure Code Generation in LLMsPreprint
Sandoval, G., Dolan-Gavitt, B., & Garg, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.16697 · Cited by 4
LiSA: Lifelong Safety Adaptation via Conservative Policy InductionPreprint
Kim, M., Miculicich, L., Mishra, B. D., Parmar, M., Wallis, P., Chandrasekhar, B., Jung, K., Pfister, T., & Le, L. T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.14454 · Cited by 4
Which Way Did It Move? Diagnosing and Overcoming Directional Motion Blindness in Video-LLMsPreprint
Lee, J., Lee, H., Kim, S., Kim, S., Chung, J., & Choi, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.22823 · Cited by 5
Voluntary Collusion with Secret Tools in Competing LLM AgentsPreprint
Zeng, X., & Rudzicz, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.27593 · Cited by 4
Alignment Collapse Under KV Cache Quantization: Diagnosis and MitigationPreprint
Xu, B. C., Kumarappan, A., & Zhou, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.09864 · Cited by 4
When is Your LLM Steerable?Preprint
Fan, C., Cheng, Y., (91180), M. L., Feizi, S., & Zhou, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.11599 · Cited by 4
Reblitz-Richardson, O. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.11375 · Cited by 4
Dual-Stance Evaluation of Sycophancy: The Structure of Agreement and the Limits of InterventionPreprint
Buchan, M. J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.11205 · Cited by 4
A Low-Rank Subspace Analysis of LLM InterventionsPreprint
Sharma, A., de Witt, C. S., Torr, P., Calinescu, A., & Yu, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.14388 · Cited by 4
Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety ReflectionPreprint
Li, J., Tang, K., Xu, Y., Ye, Z. Z., & Lyu, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.19168 · Cited by 4
FloatDoor: Platform-Triggered Backdoors in LLMsPreprint
Loose, N., Sander, J., Mächtle, F., & Eisenbarth, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.19535 · Cited by 4
HARC: Coupling Harmfulness and Refusal Directions for Robust Safety AlignmentPreprint
Chua, S. P., Wu, F., Ma, Q., & Wu, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.00572 · Cited by 4
Persona Cartography: Charting Language Model Personality Traits in Weight SpacePreprint
Baines, L., Hawthorne, A. G., Koroliuk, M., Shalibashvili, I., Dumas, C., Voudouris, K., & Africa, D. D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.07916 · Cited by 4
TRAPSBench: Vision-Language Models Encode but Fail to Express Epistemic RestraintPreprint
Pramono, F., Cai, J., & Kulkarni, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.13167 · Cited by 4
Authority Bias in Language Models: Source Deference and User Agreement Are Not InterchangeablePreprint
Kumar, A. R., & Chopra, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.37616 · Cited by 4
Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs
Sheshadri, A., Ewart, A., Guo, P., Lynch, A., Wu, C., Hebbar, V., Sleight, H., Stickland, A. C., Perez, E., Hadfield-Menell, D., et al. (2024) · arXiv · doi:10.48550/arxiv.2407.15549 · Cited by 3
Robust LLM safeguarding via refusal feature adversarial training
Yu, L., Do, V., Hambardzumyan, K., & Cancedda, N. (2024) · arXiv · doi:10.48550/arxiv.2409.20089 · Cited by 3
Quantifying the Influence of Irrelevant Contexts on Political Opinions Produced by LLMs
D’Avenia, S., & Basile, V. (2025) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2025.acl-srw.28 · Cited by 3
Greco, C. M., Cava, L. L., Zangari, L., & Tagarelli, A. (2025) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2025.acl-long.883 · Cited by 3
Toward Preference-aligned Large Language Models via Residual-based Model Steering
Cava, L. L., & Tagarelli, A. (2026) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2026/642 · Cited by 4
Progressive Behavioral Drift through Compression Valleys in Large Language Models
Xu, Z., Abaimov, S., Gardiner, J., & Belguith, S. (2026) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/2026.findings-emnlp.1967 · Cited by 3
Dong, F., Yan, Z., Ge, X., Xu, Z., Zhang, M., Chen, X., He, B., Xin, X., Chen, Z., & Zhou, Y. (2026) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/2026.emnlp-main.1453 · Cited by 3
Sperlí, G. (2025) · International Symposium on Multimedia · doi:10.1109/ism66958.2025.00011 · Cited by 3
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.