{"slug": "sedate-do-large-language-models-share-an-aggression-circuit-across-languages", "title": "SEDATE: Do Large Language Models Share an Aggression Circuit Across Languages?", "summary": "A new arXiv paper (2610.33542v1) proposes Selective Detection And Targeted Editing (SEDATE), a training-free detect-edit-verify method that identifies and edits the neurons responsible for unsafe outputs in large language models and then verifies the responses, without additional model training. The authors report that SEDATE, evaluated across 7 languages and 4 model families plus 4 out-of-distribution languages and 4 additional datasets, achieves substantially higher safety scores than the best-performing baseline in every language and model while retaining general capabilities at negligible overhead.", "body_md": "arXiv:2610.33542v1 Announce Type: new \nAbstract: As the applications of AI continue to expand across various multimodal and multilingual domains, ensuring safe and responsible deployment of these systems has become an important concern. A wide-range of interventions have been proposed to make AI responses safer and more reliable. They often incur substantial computational or annotation costs, or cannot be extended to multilingual settings. In this work, we propose Selective Detection And Targeted Editing (SEDATE), a training-free approach based on detect-edit-verify strategy. SEDATE first detects the responsible neurons, edits them as and when necessary and finally verifies the model responses; all without any additional model training. We evaluate SEDATE across 7 languages and 4 model families. We further evaluate its effectiveness across 4 out-of-distribution languages and 4 additional datasets. We show that SEDATE achieves substantially superior safety scores across all languages and models over the best-performing baseline while retaining the general capabilities and incurring negligible overhead. Our results are robust, with the trends remaining similar for out-of-distribution languages and datasets.", "url": "https://wpnews.pro/news/sedate-do-large-language-models-share-an-aggression-circuit-across-languages", "canonical_source": "https://www.machinebrief.com/news/sedate-do-large-language-models-share-an-aggression-circuit-rp2z", "published_at": "2026-10-27 04:00:00+00:00", "updated_at": "2026-10-27 04:42:23.908565+00:00", "lang": "en", "topics": ["ai-safety", "large-language-models", "ai-research", "natural-language-processing", "machine-learning"], "entities": ["SEDATE", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/sedate-do-large-language-models-share-an-aggression-circuit-across-languages", "markdown": "https://wpnews.pro/news/sedate-do-large-language-models-share-an-aggression-circuit-across-languages.md", "text": "https://wpnews.pro/news/sedate-do-large-language-models-share-an-aggression-circuit-across-languages.txt", "jsonld": "https://wpnews.pro/news/sedate-do-large-language-models-share-an-aggression-circuit-across-languages.jsonld"}}