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 801–900 of 1,000
Munigety, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.22462 · Cited by 1
Latent-space Attacks for Refusal Evasion in Language ModelsPreprint
Piras, G., Mura, R., Brau, F., Pintor, M., Oneto, L., Roli, F., & Biggio, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.21706 · Cited by 1
How Far Will They Go? Red-Teaming Online Influence with Large Language ModelsPreprint
Ruiz, D. C., Serbina, A., Rao, A., Ferrara, E., & Luceri, L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.22880 · Cited by 1
Interpretability Transfer from Language to Vision via Sparse AutoencodersPreprint
Kravets, A., Li, D., Li, C., Chen, D., & Namboodiri, V. P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.24946 · Cited by 1
Ham, S., Jang, J., Lee, W., & Kim, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.24550 · Cited by 1
Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation SimulationPreprint
Chen, L., Wang, W., Tian, Z., Zhang, C., Wu, F., Huang, J., Asiri, A., & Yu, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.24535 · Cited by 1
Palette: A Modular and Efficient Framework for On-demand Authorized Safety RelaxationPreprint
Tan, Q., Song, X., Akbari, A., Akbari, A., Wang, Y., Zhaı, X., Hong, L., Xiang, Z., Lu, J., & Yuan, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.24154 · Cited by 1
Open-Weight LLM Fine-Tuning Defenses are Susceptible to Simple AttacksPreprint
Kuo, K., Smith, V., & Yadav, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.26526 · Cited by 2
Beyond a Single Direction: Chain-of-Thought Disrupts Simple Steering of RefusalPreprint
Yang, K.-J., Meier, D., Zhao, J., Ruas, T., & Gipp, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.26772 · Cited by 2
Furina: Fragmented Uncertainty-Driven Refusal Instability AttackPreprint
Wu, T., Zhang, J., & Gao, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2605.26158 · Cited by 1
A Shared Valence Axis Across Modern LLMs and Human EEG: The Saturation RegularityPreprint
Radwan, Y. A., Liu, X., Haydarov, K., Fu, Y., & Elhoseiny, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.00129 · Cited by 1
Inference-Time Vulnerability Beyond Shallow Safety: Alignment Along Generation TrajectoriesPreprint
Park, K., & Kim, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.04778 · Cited by 1
Activation Steering of Video Generation Models via Reduced-Order Linear Optimal ControlPreprint
Hong, J., Chan, A., Dai, Q., Skifstad, J., & Chou, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.04775 · Cited by 1
Temporal Preference Concepts and their Functions in a Large Language ModelPreprint
Rios-Sialer, I., Darveshi, S., Jiang, S., Paudel, A., Pronina, A., Bandyopadhyay, I., & Shenk, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.05194 · Cited by 1
Willing but Unable: Separating Refusal from Capability in Code LLMs via AbliterationPreprint
Carleo, C., Liguori, P., Ivaki, N., & Cotroneo, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.05396 · Cited by 2
Steer Where It Matters: Token-Level Visual-Sensitivity Steering for LVLMs Hallucination MitigationPreprint
Zhang, R., Li, Z., Chen, C. L. P., & (103827), T. Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.07647 · Cited by 1
Bolliger, L. S., & Jäger, L. A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.10860 · Cited by 1
MIRAGE: A Polarity-Flipping Encoding Subspace in LLM AgentsPreprint
Revankar, P., Chauhan, K., Kim, J., Nur, S. N., Siu, V., & (27726), C. W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.10304 · Cited by 1
When the Chain of Thought Knows Better: Failure Modes in Multi-Turn Reasoning ModelsPreprint
Kasu, S. K. R., Lukas, N., & Poppi, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.10740 · Cited by 1
SafeSpec: Fast and Safe LLM via Dynamic Reflective SamplingPreprint
Xu, H., Zhang, Z., Li, L., Zheng, H., Li, Y., & Zhuo, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.19755 · Cited by 1
Leverage Is Not Reach: A Control-Window Law for Single-Neuron Steering in Language ModelsPreprint
Liu, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.19831 · Cited by 1
TIPS: Topological Ill-Posedness Probing and Steering in Large Language ModelsPreprint
Jiang, G., Tang, S., Imani, M., Bastian, N. D., & Lan, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.23590 · Cited by 1
Behavior Uncloning: Distilling Mode Redirection into Policy Weights without Inference-Time SteeringPreprint
Wang, H., Lei, J., Li, D., Liu, B., Zheng, M., Li, M., Zhang, R., & Fan, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.29201 · Cited by 1
Same Concept, Different Directions: Cross-Modal Feature Heterogeneity in Sparse AutoencodersPreprint
Lee, C., Kwon, J., Min, K., & Sohn, J.-Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.29888 · Cited by 1
EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety FailuresPreprint
Uluırmak, B. A., & Kurban, R. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2606.30219 · Cited by 1
Fast Multi-dimensional Refusal Subspaces via RFM-AGOPPreprint
Winninger, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.02396 · Cited by 2
Robust for the Wrong Reasons: The Representational Geometry of LLM Robustness to Science SkepticismPreprint
Cheon, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.01951 · Cited by 1
Language Models Represent and Transform Concepts with Shared GeometryPreprint
Hu, Z., Niu, L., & Varma, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.04525 · Cited by 2
Beyond Refusal: A Same-Lineage Study of Aligned and Abliterated LLMs for Vulnerability AnalysisPreprint
Li, M., Qiu, M., Peng, Z., Fan, H., Fu, S., Ding, J., & Feng, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.05842 · Cited by 1
Training, Reading, and Editing Legible TransformersPreprint
Oskin, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.08946 · Cited by 1
HyperSafe: Inference-Time Safety Recovery for Fine-Tuned Language ModelsPreprint
Aliev, A., Hinojosa, C. A., Eldesokey, A., An, B., Ghanem, B., & Yang, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.11475 · Cited by 1
Minionese: Comprehensive Benchmark and Mechanistic Study of Multilingual LLM SafetyPreprint
Ifebi, C., Kong, B., & Mehrotra, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.10112 · Cited by 1
Wang, W., Wang, Z., Zhan, Z., Fu, C., Li, Q., & Xu, K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.15218 · Cited by 1
Hidden in Thought: Transferable Chain-of-Thought Artifacts Induce Harmful BehaviorPreprint
Khalil, A., Kassem, A. M., Abdelrazek, M., Rana, S., Rostamzadeh, N., & Farnadi, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.15286 · Cited by 1
CircuitKIT : Circuit Discovery, Evaluation, and Application Toolkit for Mechanistic InterpretabilityPreprint
Seth, P., Gosalia, H., Kasliwal, A., & Sankarapu, V. K. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.19317 · Cited by 1
Elcock, J., Shen, W. F., Qiu, X., & Lane, N. D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.22676 · Cited by 1
Reference Feature Atlases for Mechanistic Auditing of Language ModelsPreprint
Wu, R., & Che, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.22570 · Cited by 1
OrigaMI: An LLM Serving System with First-Class Interpretability AppsPreprint
Blum, M., Silberstein, M., & David, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.22595 · Cited by 1
ToxScreen: Detecting Whether an LLM Has Been PoisonedPreprint
Hughes, A., Xing, N., Francel, C., Kim, A., & Draganov, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.26849 · Cited by 1
Tool Specifications Matter: Uncovering and Mitigating Safety Risks in AI AgentsPreprint
Pan, M., Yang, J., Yuan, Y., Jiang, Y., & Chen, Z. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.29254 · Cited by 1
Rewriting or Reweighting? A Geometric Account in Language ModelsPreprint
Wang, J., Yang, S., Wang, X., & Zhang, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.01835 · Cited by 1
SoK: Intent-Oriented Systematization of Multi-Turn LLM JailbreaksPreprint
Li, S., Wulianghai, A., Liu, Z., Lin, X., Mao, Q., Li, H., Chen, X., Liang, S., Wu, J., Li, J., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.01117 · Cited by 1
Khemais, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.03620 · Cited by 1
Reliability-Safety Trade-off in AI Distillation: A Renormalization-Group ApproachPreprint
Du, Y. M., Yi, M.-M., Zhou, T.-J., & Sun, C. P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.08572 · Cited by 1
The Knowing-Saying Gap: When Probes See Errors that Confidence MissesPreprint
Goel, J., Bandyopadhyay, I., & Shenk, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.07528 · Cited by 1
Cao, L. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.11705 · Cited by 1
A Probe Direction Is a Property of Its PromptPreprint
Noël, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.13329 · Cited by 1
Nine Emotion Centroids: A Label-Free Valence Axis That Transfers Across Four ModalitiesPreprint
Radwan, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.18090 · Cited by 1
Hidden in the Request: Explaining Unethical LLM Compliance through Token RelevancePreprint
Biton, O., Krichli, T., Allouche, I., & Keshet, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.23264 · Cited by 1
Reading Is Not Using: Retrieval, Judgment, and the Design of AI Financial Research WorkflowsPreprint
Liu, M., & Liu, Z. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7349358 · Cited by 1
How Language Models Organize and Structure Moral KnowledgePreprint
Reblitz-Richardson, O. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.27402 · Cited by 2
Chowdhury, M. M., Chang, E., & Li, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.30585 · Cited by 1
Context-Grounding Gains Are Mediated by Pre-existing Machinery: Auditing GRPO, SFT, and DPOPreprint
Gupta, P., & Gupta, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.00925 · Cited by 1
The Geometry of Refusal: Why Post-Hoc Refusal Is Fragile and Pretraining-Time Refusal PersistsPreprint
Malla, S., Choi, C., & Choi, J. H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.06934 · Cited by 1
Capsule Lens: Locating and Tracking Concept Geometry in Model RepresentationsPreprint
Tang, Y., Saini, H., Jha, S., Chen, H., Duan, X., & Liu, D. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.05575 · Cited by 1
Steering Interference Reflects the Model's Defaults, Not the Behavior DirectionsPreprint
Malla, S., Choi, C., & Choi, J. H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.06951 · Cited by 1
Disentangling Steering VectorsPreprint
Hiramatsu, T., Atarashi, K., Takeuchi, K., & Kashima, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.07037 · Cited by 1
Interpreting and Steering LLM Agents for Social SimulationsPreprint
Fan, J. G., Murugan, A., Krishnan, S., & Nagaraj, A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.16436 · Cited by 1
Test-Time Unlearning via Sparse AutoencoderPreprint
Li, P., Duan, J., Tadiparthi, V., Agarwal, N., Lee, K., Moradi‐Pari, E., Mahjoub, H. N., Liu, S., & Chen, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.16229 · Cited by 1
Decoy Direction Optimization: A Post-Hoc Defense Against LLM AbliterationPreprint
Muhamed, A., Diab, M., & Smith, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.16204 · Cited by 2
Aamir, S., & Adil, M. A. B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.17550 · Cited by 1
Xeno-Interpretability: Investigating the Alien Minds of LLMsPreprint
Pierucci, F., Syrnikov, M. B., Prandi, M., Galisai, M., Giarrusso, F., & Bisconti, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.20408 · Cited by 1
Same Outcome, Different Readout: What Does a Steerable Valence Direction in LLMs Represent?Preprint
Li, W., Chen, X., Song, Y., Lin, X., & Zheng, T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.22850 · Cited by 1
Knowing, and Saying It Only When Asked: LLM Endognostics and the Schizognosis of Minerva-7BPreprint
Davide, F., & Collovà, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.22219 · Cited by 1
Slow Decay and Silenced Expression: Iterated Subliminal Trait Transfer in Language-Model LineagesPreprint
Vo, R., Nguyen, D.-V., Kretchmar, M., & Nguyen, N. L.-T. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.25721 · Cited by 1
Refusing Everything Looks Safe: Restoring the Benign Arm to Encoded-Prompt EvaluationPreprint
Zhang, H., Xu, H., Luo, X., Liu, H., Chen, Y., Xiao, Z., Feng, Y., Guan, X., Zandsalimy, M., & Sushmita, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.26176 · Cited by 1
Matryoshka attribution: Learning to attribute language model outputs to representations and weightsPreprint
Arora, A., Acharya, K., Hu, N., Zhang, Y., Goodman, N. D., Jurafsky, D., & Potts, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.25518 · Cited by 1
Tool Mediation Alters Refusal Mechanisms in Large Language ModelsPreprint
Rodríguez, A., Garofalo, G., Desmet, L., & Rimmer, V. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.35117 · Cited by 1
Steering Language Model Goals with Value TransplantPreprint
Jiang, P., & Roger, F. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.34056 · Cited by 1
Says Block, Still Acts: Why LLM Safety Judgments Fail to Govern Action in LLM AgentsPreprint
Chen, D., Zhao, X., Yao, X., & Wei, X. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.35870 · Cited by 1
Alignment via Training Against Probes Without Losing MonitorabilityPreprint
Libon, L., Panfilov, A., Rank, B., Chen, X., Geiping, J., & Andriushchenko, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.38645 · Cited by 1
Refusal Localizes, the Damage Relocates: Safety Layers Under Few-Sample Fine-TuningPreprint
Lee, J., Lee, D. J., Eo, S., Hong, S., Lee, S., & Lim, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.00320 · Cited by 1
SafeDepth: Safety-Aware Token-Level Adaptive ComputationPreprint
Li, N., & Harris, I. G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.00815 · Cited by 1
Representation Transitions Reveal Emerging Safety Risks in Multi-Turn LLM AgentsPreprint
Wang, H., Zhao, W., Zhang, Y., POSKITT, C. M., & Sun, J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.00400 · Cited by 1
Kernelized Activation SteeringPreprint
Abdullaev, L. U., Pham, M., Do, B., Than, K., & Nguyen, T. M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.01062 · Cited by 1
On the Steering Dimensionality of Refusal in Language ModelsPreprint
Wang, H., Miehling, E., Wei, D., Ramamurthy, K. N., & Zhang, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.04245 · Cited by 2
Hashmi, S., Ranjan, M., Elsayed, A., Sheikh, M. U., Shamshad, F., & Lukas, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.05894 · Cited by 1
Base Models Can Reason By Taking a Cue From Training DataPreprint
Wang, S. L., Dravid, A. V., Shao, R., Farhat, K., Min, S., & Efros, A. A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.06851 · Cited by 1
Karamat, M. Z., & Garcia, C. C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.09000 · Cited by 1
From Local Evidence to Safety Verdicts: Causal Tracing in Vision-Language ModelsPreprint
Tarzjani, F. D., Wijewardena, M., Romanus, A., & Mohanty, S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.07514 · Cited by 1
The Persona Hierarchy Model: Understanding Contextual Generalization in Fine-Tuning LLMsPreprint
Zhao, J., Wu, Z., Bau, D., & Shi, W. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.09384 · Cited by 1
Does the Model Use the Feature? Separating Steering from Mechanism in LLMsPreprint
Che, T., & Li, Y. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.07270 · Cited by 1
Visual Memory Attacks Can Persist Through The KV CachePreprint
Dobre, D., Schwinn, L., Gidel, G., Gella, S., Taslakian, P., & Noël, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.09027 · Cited by 1
Telling FIBS: Probes Effectively Detect Sabotage and Catch Unverbalized DeceptionPreprint
Hollinsworth, O. J., Spies, A. F., Diriba, T., Gleave, A., & Cundy, C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.12445 · Cited by 1
Ghost Robots: What the Law Can Do When No One Is Behind the Machine
Gilly, T. (2026) · doi:10.68481/8qa2p9 · Cited by 1
Are You Reading the Right Thing? Rethinking Internal Signals for Prompt-Injection DetectionPreprint
Gupta, S., & Tembey, P. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.19452 · Cited by 1
Comparing Sparsified MLPs and Transcoders: Many Transcoder Features Are Neurons in DisguisePreprint
Shabalin, S., & Piotrowski, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.24521 · Cited by 1
Dr. No: Feature Steering for Abstention in Vision-Language ModelsPreprint
Hasanov, F., Renzulli, R., Zonooz, B., Pechenizkiy, M., & Mocanu, D. C. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2611.03862 · Cited by 1
Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
Vassilev, A. T. (2026) · doi:10.6028/nist.sp.2400-102e2026.ipd · Cited by 1
Ahmed, N. S. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2611.08070 · Cited by 1
Shaping & Monitoring Models through their Internal RepresentationsPreprint
Fel, T., Kowal, M. D., Feiszli, M., Grant, S., Lee, A., Earls, C., Dinç, F., Shyam, V., Bhalla, U., Bergen, L., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2611.07071 · Cited by 1
Understanding AI from the Inside: The State of Mechanistic Interpretability for AI SafetyPreprint
Barez, F., Wu, T.-Y., Batra, H., Dominici, G., Draye, F., Elazar, Y., Fornasiere, D., García, A., Harrasse, A., Harris, R., et al. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2611.08438 · Cited by 1
Model Organisms for Emergent Misalignment
Turner, E., Soligo, A., Taylor, M., Rajamanoharan, S., & Nanda, N. (2025) · arXiv · doi:10.48550/arxiv.2506.11613
Self-citationConvergent Linear Representations of Emergent Misalignment
Soligo, A., Turner, E., Rajamanoharan, S., & Nanda, N. (2025) · arXiv · doi:10.48550/arxiv.2506.11618
Self-citationZhang, N., Kwek, E., Zhang, Y., Nguyen, N.-H., Mitra, P., & Zhang, R. (2025) · arXiv · doi:10.48550/arxiv.2504.02010
Instruction Following by Principled Boosting Attention of Large Language Models
Guardieiro, V., Khare, A., Stein, A., & Wong, E. (2025) · arXiv · doi:10.48550/arxiv.2506.13734
The Value Axis: Language Models Encode Whether They're on the Right Track
Jiang, N., Kauvar, I., & Lindsey, J. (2026) · arXiv · doi:10.48550/arxiv.2606.17056
The Logical Implication Steering Method for Conditional Interventions on Transformer GenerationPreprint
Kalajdzievski, D. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2502.03618
Bajwa, T. S. (2025) · Communications in computer and information science · doi:10.1007/978-3-031-98949-0_10
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.