{"id":"https://openalex.org/W7204875884","doi":"https://doi.org/10.18653/2026.findings-emnlp.1840","title":"The Fragility of Jailbreak Robustness Across Operational States","display_name":"The Fragility of Jailbreak Robustness Across Operational States","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7204875884","doi":"https://doi.org/10.18653/2026.findings-emnlp.1840"},"language":"en","primary_location":{"id":"doi:10.18653/2026.findings-emnlp.1840","doi":"https://doi.org/10.18653/2026.findings-emnlp.1840","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.findings-emnlp.1840","pdf_url":"https://aclanthology.org/2026.findings-emnlp.1840.pdf","source":{"id":"https://openalex.org/S7407096085","display_name":"Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2026","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://aclanthology.org/2026.findings-emnlp.1840.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137569473","display_name":"Yuna Park","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]},{"id":"https://openalex.org/I58716616","display_name":"Korea Institute of Science and Technology","ror":"https://ror.org/05kzfa883","country_code":"KR","type":"facility","lineage":["https://openalex.org/I27494661","https://openalex.org/I2801339556","https://openalex.org/I4210144908","https://openalex.org/I4387152098","https://openalex.org/I58716616"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yuna Park","raw_affiliation_strings":["Korea Institute of Science and Technology (KIST)","Yonsei University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Institute of Science and Technology (KIST)","institution_ids":["https://openalex.org/I58716616"]},{"raw_affiliation_string":"Yonsei University","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052904936","display_name":"Hwang Youn Kim","orcid":"https://orcid.org/0000-0003-0181-5693","observed_orcids":["https://orcid.org/0000-0003-0181-5693"]},"institutions":[{"id":"https://openalex.org/I58716616","display_name":"Korea Institute of Science and Technology","ror":"https://ror.org/05kzfa883","country_code":"KR","type":"facility","lineage":["https://openalex.org/I27494661","https://openalex.org/I2801339556","https://openalex.org/I4210144908","https://openalex.org/I4387152098","https://openalex.org/I58716616"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hwang Youn Kim","raw_affiliation_strings":["Korea Institute of Science and Technology (KIST)"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Institute of Science and Technology (KIST)","institution_ids":["https://openalex.org/I58716616"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5148908821","display_name":"Yujin Kim","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I197347611","display_name":"Korea University","ror":"https://ror.org/047dqcg40","country_code":"KR","type":"education","lineage":["https://openalex.org/I197347611"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yujin Kim","raw_affiliation_strings":["Korea University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea University","institution_ids":["https://openalex.org/I197347611"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017913155","display_name":"Won Woo Ro","orcid":"https://orcid.org/0000-0001-5390-6445","observed_orcids":["https://orcid.org/0000-0001-5390-6445"]},"institutions":[{"id":"https://openalex.org/I193775966","display_name":"Yonsei University","ror":"https://ror.org/01wjejq96","country_code":"KR","type":"education","lineage":["https://openalex.org/I193775966"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Won Woo Ro","raw_affiliation_strings":["Yonsei University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Yonsei University","institution_ids":["https://openalex.org/I193775966"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5159954820","display_name":"Suhyun Kim","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I35928602","display_name":"Kyung Hee University","ror":"https://ror.org/01zqcg218","country_code":"KR","type":"education","lineage":["https://openalex.org/I35928602"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Suhyun Kim","raw_affiliation_strings":["Kyung Hee University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Kyung Hee University","institution_ids":["https://openalex.org/I35928602"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5028302461","display_name":"Jae\u2010In Hwang","orcid":"https://orcid.org/0000-0002-8748-9077","observed_orcids":["https://orcid.org/0000-0002-8748-9077"]},"institutions":[{"id":"https://openalex.org/I58716616","display_name":"Korea Institute of Science and Technology","ror":"https://ror.org/05kzfa883","country_code":"KR","type":"facility","lineage":["https://openalex.org/I27494661","https://openalex.org/I2801339556","https://openalex.org/I4210144908","https://openalex.org/I4387152098","https://openalex.org/I58716616"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jae-In Hwang","raw_affiliation_strings":["Korea Institute of Science and Technology (KIST)"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Institute of Science and Technology (KIST)","institution_ids":["https://openalex.org/I58716616"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.68804575,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"39277","last_page":"39298"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9994999766349792,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9994999766349792,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11424","display_name":"Security and Verification in Computing","score":9.999999747378752e-05,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7239999771118164},{"id":"https://openalex.org/keywords/operating-states","display_name":"operating states","score":0.7919999957084656},{"id":"https://openalex.org/keywords/system-prompts","display_name":"system prompts","score":0.6890000104904175},{"id":"https://openalex.org/keywords/attack-success-rate","display_name":"attack success rate","score":0.8999999761581421},{"id":"https://openalex.org/keywords/refusal-behavior","display_name":"refusal behavior","score":0.7400000095367432},{"id":"https://openalex.org/keywords/hidden-representations","display_name":"hidden representations","score":0.9739999771118164},{"id":"https://openalex.org/keywords/robustness-analysis","display_name":"robustness analysis","score":0.675000011920929},{"id":"https://openalex.org/keywords/prompt-engineering","display_name":"prompt engineering","score":0.4050000011920929}],"concepts":[{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.8812000155448914},{"id":"https://openalex.org/C80191262","wikidata":"https://www.wikidata.org/wiki/Q5477668","display_name":"Fragility","level":2,"score":0.7853000164031982},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5742999911308289},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.37040001153945923},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.30169999599456787},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.26600000262260437},{"id":"https://openalex.org/C2776544517","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Unexpected events","level":2,"score":0.2502000033855438},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.24300000071525574},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2363000065088272},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.22339999675750732}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/2026.findings-emnlp.1840","doi":"https://doi.org/10.18653/2026.findings-emnlp.1840","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.findings-emnlp.1840","pdf_url":"https://aclanthology.org/2026.findings-emnlp.1840.pdf","source":{"id":"https://openalex.org/S7407096085","display_name":"Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2026","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2608.30748","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.30748","pdf_url":"https://arxiv.org/pdf/2608.30748","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.30748","doi":"https://doi.org/10.48550/arxiv.2608.30748","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.30748","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/2026.findings-emnlp.1840","doi":"https://doi.org/10.18653/2026.findings-emnlp.1840","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.findings-emnlp.1840","pdf_url":"https://aclanthology.org/2026.findings-emnlp.1840.pdf","source":{"id":"https://openalex.org/S7407096085","display_name":"Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1370073529","display_name":null,"funder_award_id":"IITP-2026-RS-2023-00258649","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G1839198279","display_name":null,"funder_award_id":"IITP-2026-RS-2023-00258649","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"},{"id":"https://openalex.org/G3621670845","display_name":null,"funder_award_id":"26E0212","funder_id":"https://openalex.org/F4320322091","funder_display_name":"Korea Institute of Science and Technology"},{"id":"https://openalex.org/G3812690899","display_name":null,"funder_award_id":"RS-2023","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"}],"funders":[{"id":"https://openalex.org/F4320322091","display_name":"Korea Institute of Science and Technology","ror":"https://ror.org/05kzfa883"},{"id":"https://openalex.org/F4320324891","display_name":"Iran Telecommunication Research Center","ror":"https://ror.org/01a3g2z22"},{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204875884.pdf","grobid_xml":"https://content.openalex.org/works/W7204875884.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4415795355","https://openalex.org/W7119531022","https://openalex.org/W7160898367","https://openalex.org/W7223054139","https://openalex.org/W7220748425","https://openalex.org/W4414450005","https://openalex.org/W7206196466","https://openalex.org/W4416100781","https://openalex.org/W4399695855","https://openalex.org/W7161030552"],"abstract_inverted_index":{"Existing":[0],"jailbreak":[1,37,79,150,163],"evaluations":[2,166],"typically":[3],"characterize":[4,162],"robustness":[5,38,128,171],"using":[6],"a":[7,15,106,139,155],"single":[8,156],"attack":[9,48,65],"success":[10,66],"rate":[11],"(ASR)":[12],"measured":[13],"in":[14,84,108,135],"default":[16],"configuration":[17],"(the":[18],"vanilla":[19,30,87],"state).However,":[20],"user-LLM":[21],"interactions":[22],"can":[23,62],"induce":[24],"diverse":[25],"operational":[26,90,109,175],"states":[27],"beyond":[28],"the":[29,47],"state.In":[31],"this":[32,70,146],"work,":[33],"we":[34],"find":[35],"that":[36,126,143,154,167],"is":[39,130],"highly":[40],"fragile":[41],"to":[42,59,98,105],"operational-state":[43],"variation:":[44],"even":[45,114],"when":[46],"remains":[49],"fixed,":[50],"changing":[51],"only":[52],"an":[53],"ordinary":[54],"system":[55],"prompt":[56],"not":[57,160],"designed":[58,118],"affect":[60],"safety":[61],"dramatically":[63],"alter":[64],"rates.We":[67],"systematically":[68,131],"investigate":[69],"phenomenon":[71],"across":[72,173],"seven":[73],"aligned":[74],"models":[75],"and":[76,88,119,142],"three":[77],"representative":[78],"attacks,":[80],"observing":[81],"substantial":[82],"differences":[83,134],"ASR":[85,94],"between":[86],"non-vanilla":[89,174],"states.In":[91],"one":[92],"case,":[93],"increases":[95,112],"by":[96],"up":[97],"56":[99],"percentage":[100],"points":[101],"(2%\u219258%)":[102],"solely":[103],"due":[104],"change":[107],"state.Remarkably,":[110],"these":[111],"occur":[113],"for":[115],"attacks":[116],"originally":[117],"optimized":[120],"under":[121],"vanilla-state":[122,157],"evaluation.We":[123],"further":[124],"show":[125,153],"state-dependent":[127],"variation":[129],"associated":[132],"with":[133],"hidden":[136],"representations":[137],"along":[138],"refusal-related":[140],"axis,":[141],"projections":[144],"onto":[145],"axis":[147],"strongly":[148],"predict":[149],"outcomes.Our":[151],"results":[152],"evaluation":[158],"may":[159],"fully":[161],"robustness,":[164],"motivating":[165],"also":[168],"examine":[169],"how":[170],"changes":[172],"states.":[176],"jailbreak-fragility-across-states":[177]},"counts_by_year":[],"updated_date":"2026-11-01T11:40:01.658920","created_date":"2026-09-02T00:00:00"}