[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-emergent-misalignment":3},{"tag":4,"articles":10},{"id":5,"name":6,"slug":7,"article_count":8,"description_zh":9,"description_en":9},"e339d1d8-5adb-46e8-957c-ecb0e9e0240f","emergent misalignment","emergent-misalignment",2,null,[11],{"id":12,"slug":13,"title":14,"summary":15,"category":16,"image_url":17,"cover_image":17,"language":18,"created_at":19},"4c790996-8883-4535-ac50-350e7882f0a3","llms-harmful-content-unified-mechanism-en","How LLMs encode harmful behavior internally","A pruning study suggests harmful output lives in a compact, shared weight set—helping explain jailbreak brittleness and emergent misalignment.","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776060239898-zqk2.png","en","2026-04-13T06:03:38.491469+00:00"]