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
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.