phenomenon · first seen 7 Oct 2026
refusal behaviour
Also: safety refusal, refusal steering, refusal direction
Language models declining harmful requests, which can be localized to specific directions in internal activations.
Mentions per week
PapersCodeDiscussion
Earliest sources
- How Narrative Wrapping Affects LLM Refusal: A Cross-Language Benchmark and DefensePapers · 7 Oct 2026 · Zhankai Ye, Yanning Wang, Yukai Jin, Bo Mei, Fangyi Li, Wei Wang, Shangqian Gao,…
- Running Fewer Loops Removes Refusal Behaviour in Recurrent-Depth Language ModelsPapers · 10 Oct 2026 · Seyedarmin Azizi, Erfan Baghaei Potraghloo, Arya Fayyazi, Saeid Shokoufa, Massou…
- Risk and Refusal: Mechanisms of LLM Responses to Suicidal IdeationPapers · 15 Oct 2026 · Drishti Goel, Soorya Ram Shimgekar, Jeongah Lee, Violeta J. Rodriguez, Daniel S.…
- Is Refusal Direction Universal across Low-Resourced Languages?Papers · 20 Oct 2026 · Anastasiia Ivanova
- WRDO: Weighted Refusal Direction Optimization in Aligned Language ModelsPapers · 21 Oct 2026 · Grigoriy Evseev, Vladimir Solodkin, Dmitry Bylinkin, Daniil Medyakov, Vladimir A…
- Depth-Dependent Safety Geometry in Vision-Language Models: Layer-Wise Localization and Bidirectional Refusal ControlPapers · 23 Oct 2026 · Muhammad Zeeshan Karamat, Syed Talal Hassan, Christiana Chamon Garcia
Latest sources
- Abstention Direction in Large Language Models: Mechanistic Interpretability of Abstention Across Query Ambiguity, Knowledge Limits, and Safety AlignmentPapers · 16 Nov 2026 · Koki Itai, Jun Hirako, Hiroaki Yamagiwa
- Refusal Register is Causally Steerable: Decomposing the Refusal Direction into Category-Aligned Components in Gemma-2-9B-ITPapers · 16 Nov 2026 · Janhavi Khindkar, Nataliia Povarova
- Centered Ablation: Why Last-Layer Directional Ablation Under-Fires or Collapses, and a Closed-Form FixPapers · 15 Nov 2026 · Quy-Anh Dang, Chris Ngo
- Where Is Rank-1 DiM Refusal Intervenable in Multimodal LM Derivatives?Papers · 15 Nov 2026 · Chansu Han, Seok-Hwan Choi, Takeshi Takahashi
- Is Abliteration Getting Harder? Not Uniformly: Refusal Removal by Direction-Based Ablation and ARA in Heretic on the Qwen3.8, Gemma 4, Olmo 3.1 and Mistral Small 3.2 LLMs Versus Their PredecessorsPapers · 13 Nov 2026 · Richard J. Young
- Interpretability Is Not Model-SpecificPapers · 9 Nov 2026 · Marcus Armstrong, Navid Ayoobi, Pradham Mummaleti, Alexander Chulzhanov, Arjun M…
- Restoring General and Improving Domain-Specific Safety in Fine-Tuned LLMs Without RetrainingPapers · 6 Nov 2026 · Vishnu Kabir Chhabra, Ding Zhu, Zhihui Zhu, Xueru Zhang, Mohammad Mahdi Khalili
- Compositional Unalignment: Auditing Safety in Large Vision-Language ModelsPapers · 4 Nov 2026 · Hong kyu Lee, Kai Shu, Li Xiong
- Which Words a Model Fears: From a Refusal Direction to a Vocabulary-Wide Fear LexiconPapers · 2 Nov 2026 · Boyuan Zhang, Mohammad Zandsalimy, Shanu Sushmita
- J-Anchor: Defending Against Prefill Attacks via J-Space Selective Refusal RestorationPapers · 31 Oct 2026 · Running Yang, Beidi Zhao, Xiaoxiao Li
- Depth-Dependent Safety Geometry in Vision-Language Models: Layer-Wise Localization and Bidirectional Refusal ControlPapers · 23 Oct 2026 · Muhammad Zeeshan Karamat, Syed Talal Hassan, Christiana Chamon Garcia
- WRDO: Weighted Refusal Direction Optimization in Aligned Language ModelsPapers · 21 Oct 2026 · Grigoriy Evseev, Vladimir Solodkin, Dmitry Bylinkin, Daniil Medyakov, Vladimir A…