{"id":"https://openalex.org/W4412889922","doi":"https://doi.org/10.18653/v1/2025.acl-long.1233","title":"Shaping the Safety Boundaries: Understanding and Defending Against Jailbreaks in Large Language Models","display_name":"Shaping the Safety Boundaries: Understanding and Defending Against Jailbreaks in Large Language Models","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412889922","doi":"https://doi.org/10.18653/v1/2025.acl-long.1233"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.1233","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1233","pdf_url":"https://aclanthology.org/2025.acl-long.1233.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.acl-long.1233.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119929742","display_name":"Lang Gao","orcid":"https://orcid.org/0000-0002-6607-3240"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lang Gao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073054282","display_name":"Jiahui Geng","orcid":"https://orcid.org/0000-0002-4205-8230"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiahui Geng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000755750","display_name":"Xiangliang Zhang","orcid":"https://orcid.org/0000-0002-3574-5665"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiangliang Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012055259","display_name":"Preslav Nakov","orcid":"https://orcid.org/0000-0002-3600-1510"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Preslav Nakov","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5042441263","display_name":"Xiuying Chen","orcid":"https://orcid.org/0000-0002-6633-0796"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiuying Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":30.1621,"has_fulltext":true,"cited_by_count":5,"citation_normalized_percentile":{"value":0.99290399,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":98,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"25378","last_page":"25398"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T13643","display_name":"Artificial Intelligence in Law","score":0.8949000239372253,"subfield":{"id":"https://openalex.org/subfields/3320","display_name":"Political Science and International Relations"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T13643","display_name":"Artificial Intelligence in Law","score":0.8949000239372253,"subfield":{"id":"https://openalex.org/subfields/3320","display_name":"Political Science and International Relations"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.864300012588501,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T12921","display_name":"European Criminal Justice and Data Protection","score":0.8087000250816345,"subfield":{"id":"https://openalex.org/subfields/3320","display_name":"Political Science and International Relations"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6452126502990723},{"id":"https://openalex.org/keywords/linguistics","display_name":"Linguistics","score":0.375896155834198},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.3541693091392517},{"id":"https://openalex.org/keywords/human\u2013computer-interaction","display_name":"Human\u2013computer interaction","score":0.3535637855529785},{"id":"https://openalex.org/keywords/programming-language","display_name":"Programming language","score":0.34807467460632324},{"id":"https://openalex.org/keywords/cognitive-science","display_name":"Cognitive science","score":0.34176141023635864},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.21418455243110657},{"id":"https://openalex.org/keywords/philosophy","display_name":"Philosophy","score":0.12724938988685608}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6452126502990723},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.375896155834198},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3541693091392517},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.3535637855529785},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.34807467460632324},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.34176141023635864},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.21418455243110657},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.12724938988685608}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.1233","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1233","pdf_url":"https://aclanthology.org/2025.acl-long.1233.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.acl-long.1233","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1233","pdf_url":"https://aclanthology.org/2025.acl-long.1233.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.5299999713897705,"id":"https://metadata.un.org/sdg/16","display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412889922.pdf","grobid_xml":"https://content.openalex.org/works/W4412889922.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4231937131","https://openalex.org/W323219885","https://openalex.org/W2063928587","https://openalex.org/W1487966966","https://openalex.org/W1589342014","https://openalex.org/W1480341462","https://openalex.org/W598950423","https://openalex.org/W4256673449","https://openalex.org/W4292559016","https://openalex.org/W2143331802"],"abstract_inverted_index":{"Jailbreaking":[0],"in":[1,48,87],"large":[2],"language":[3],"models":[4,12],"(LLMs)":[5],"poses":[6],"major":[7],"security":[8],"risks":[9],"by":[10],"tricking":[11],"into":[13],"generating":[14],"harmful":[15,61,74],"text.However,":[16],"there":[17],"is":[18],"limited":[19],"understanding":[20],"of":[21,40,165],"how":[22],"jailbreaking":[23],"operates,":[24],"which":[25,111],"makes":[26],"it":[27],"difficult":[28],"to":[29,73,126],"develop":[30],"effective":[31,130],"defenses.In":[32],"this":[33],"work,":[34],"we":[35,100,122],"conduct":[36],"a":[37,64,95,102],"large-scale":[38],"analysis":[39,56],"seven":[41],"jailbreak":[42,170],"methods":[43],"and":[44,81,135,140],"uncover":[45],"that":[46,58,78,137,156],"inconsistencies":[47],"previous":[49],"studies":[50],"arise":[51],"from":[52],"insufficient":[53],"observation":[54],"samples.Our":[55],"reveals":[57],"jailbreaks":[59],"shift":[60],"activations":[62,114],"beyond":[63],"defined":[65],"safety":[66,117],"boundary,":[67],"where":[68],"LLMs":[69],"become":[70],"less":[71,173],"sensitive":[72],"information.We":[75],"also":[76],"find":[77],"the":[79,82,116,128,138,144,178],"low":[80,139],"middle":[83,141],"layers":[84,93,131,142],"are":[85],"critical":[86],"driving":[88],"these":[89,98],"shifts,":[90],"while":[91],"deeper":[92],"play":[94],"lesser":[96],"role.Leveraging":[97],"insights,":[99],"propose":[101],"novel":[103],"defense":[104,161],"mechanism":[105],"called":[106],"Activation":[107],"Boundary":[108],"Defense":[109],"(ABD),":[110],"adaptively":[112],"constrains":[113],"within":[115],"boundary.To":[118],"further":[119],"optimize":[120],"performance,":[121],"use":[123],"Bayesian":[124],"optimization":[125],"select":[127],"most":[129],"for":[132],"ABD":[133,157],"application":[134],"confirm":[136],"have":[143],"greatest":[145],"impact,":[146],"consistent":[147],"with":[148,172],"our":[149],"earlier":[150],"observations.Experiments":[151],"across":[152],"multiple":[153],"benchmarks":[154],"demonstrate":[155],"achieves":[158],"an":[159],"average":[160],"success":[162],"rate":[163],"(DSR)":[164],"over":[166],"98%":[167],"against":[168],"various":[169],"attacks,":[171],"than":[174],"2%":[175],"impact":[176],"on":[177],"model's":[179],"overall":[180],"capabilities.":[181]},"counts_by_year":[{"year":2026,"cited_by_count":5}],"updated_date":"2026-08-08T01:25:22.217667","created_date":"2025-10-10T00:00:00"}